update readme

README.md

@@ -18,7 +18,6 @@ tags:
 - pointcloud
 - multimodal
 ---
-
 <div align="center">
     <h1 align="center">The P<sup>3</sup> dataset: Pixels, Points and Polygons <br> for Multimodal Building Vectorization</h1>
     <h3><align="center">Raphael Sulzer<sup>1,2</sup> &nbsp;&nbsp;&nbsp; Liuyun Duan<sup>1</sup>
@@ -508,7 +507,6 @@ pip install .
 ⚠️ **Warning**: The implementation of the LiDAR point cloud encoder uses Open3D-ML. Currently, Open3D-ML officially only supports the PyTorch version specified in `requirements-torch-cuda.txt`.
 
 
-
 <!-- ## Model Zoo
 
 
@@ -550,13 +548,16 @@ python scripts/train.py --help
 
 </details> -->
 
-### Predict a single tile
+### Predict demo tile
 
-TODO
+After downloading the model weights and setting up the code you can predict a demo tile by running
 
 ```
-python scripts/predict_demo.py
+python scripts/predict_demo.py checkpoint=best_val_iou experiment=$MODEL_$MODALITY +image_file=demo_data/image0_CH_val.tif +lidar_file=demo_data/lidar0_CH_val.copc.laz
 ```
+At least one of `image_file` or `lidar_file` has to be specified. `$MODEL` can be one of the following: `ffl`, `hisup` or `p2p`. `$MODALITY` can be `image`, `lidar` or `fusion`.
+The result will be stored in `prediction.png`.
+
 
 ### Reproduce paper results
 
@@ -592,7 +593,7 @@ python scripts/train.py experiment=p2p_fusion checkpoint=latest
 
 ## Citation
 
-If you find our work useful, please consider citing:
+If you use our work please cite
 ```bibtex
 TODO
 ```

ffl/224/v0_all_bs4x16/.hydra/config.yaml

@@ -1,117 +1,32 @@
 host:
-  name: jeanzay
-  data_root: /lustre/fswork/projects/rech/cso/uku93eu/data
-  update_pbar_every: 60
+  name: gin
+  data_root: /data/rsulzer/${..dataset.name}
+  model_root: /data/rsulzer/${..dataset.name}_output
+  multi_gpu: false
+  device: cuda
+  update_pbar_every: 1
+  ldof_exe: /user/rsulzer/home/cpp/line-DOF-metric/build/calculate_DoF
 run_type:
-  name: release
+  name: debug
   batch_size: 16
-  train_subset: null
-  val_subset: null
-  test_subset: null
-  logging: INFO
-  num_workers: 16
-  log_to_wandb: true
-polygonization:
-  method:
-  - acm
-  common_params:
-    init_data_level: 0.5
-  simple_method:
-    data_level: 0.5
-    tolerance:
-    - 1.0
-    seg_threshold: 0.5
-    min_area: 10
-  asm_method:
-    init_method: skeleton
-    data_level: 0.5
-    loss_params:
-      coefs:
-        step_thresholds:
-        - 0
-        - 100
-        - 200
-        - 300
-        data:
-        - 1.0
-        - 0.1
-        - 0.0
-        - 0.0
-        crossfield:
-        - 0.0
-        - 0.05
-        - 0.0
-        - 0.0
-        length:
-        - 0.1
-        - 0.01
-        - 0.0
-        - 0.0
-        curvature:
-        - 0.0
-        - 0.0
-        - 1.0
-        - 0.0
-        corner:
-        - 0.0
-        - 0.0
-        - 0.5
-        - 0.0
-        junction:
-        - 0.0
-        - 0.0
-        - 0.5
-        - 0.0
-      curvature_dissimilarity_threshold: 2
-      corner_angles:
-      - 45
-      - 90
-      - 135
-      corner_angle_threshold: 22.5
-      junction_angles:
-      - 0
-      - 45
-      - 90
-      - 135
-      junction_angle_weights:
-      - 1
-      - 0.01
-      - 0.1
-      - 0.01
-      junction_angle_threshold: 22.5
-    lr: 0.1
-    gamma: 0.995
-    device: cuda
-    tolerance:
-    - 1
-    seg_threshold: 0.5
-    min_area: 10
-  acm_method:
-    steps: 500
-    data_level: 0.5
-    data_coef: 0.1
-    length_coef: 0.4
-    crossfield_coef: 0.5
-    poly_lr: 0.01
-    warmup_iters: 100
-    warmup_factor: 0.1
-    device: cuda
-    tolerance:
-    - 1
-    seg_threshold: 0.5
-    min_area: 10
+  train_subset: 256
+  val_subset: 32
+  test_subset: 32
+  logging: DEBUG
+  num_workers: 0
+  log_to_wandb: false
 dataset:
-  name: lidarpoly
+  name: PixelsPointsPolygons
   size: ${..experiment.encoder.in_size}
-  path: ${host.data_root}/${.name}/${.size}
+  path: ${host.data_root}/data/${.size}
   annotations:
-    train: ${..path}/annotations_${...country}_train.json
-    val: ${..path}/annotations_${...country}_val.json
-    test: ${..path}/annotations_${...country}_test.json
+    train: ${..path}/annotations/annotations_${...experiment.country}_train.json
+    val: ${..path}/annotations/annotations_${...experiment.country}_val.json
+    test: ${..path}/annotations/annotations_${...experiment.country}_test.json
   ffl_stats:
-    train: ${..path}/ffl/train/stats-${...country}.pt
-    val: ${..path}/ffl/val/stats-${...country}.pt
-    test: ${..path}/ffl/test/stats-${...country}.pt
+    train: ${..path}/ffl/train/stats-${...experiment.country}.pt
+    val: ${..path}/ffl/val/stats-${...experiment.country}.pt
+    test: ${..path}/ffl/test/stats-${...experiment.country}.pt
   train_subset: ${..run_type.train_subset}
   val_subset: ${..run_type.val_subset}
   test_subset: ${..run_type.test_subset}
@@ -137,7 +52,7 @@ experiment:
     out_feature_dim: ${..model.decoder.in_feature_dim}
     vit:
       type: vit_small_patch${..patch_size}_${..in_size}.dino
-      checkpoint_file: ${....host.data_root}/checkpoints/backbones/dino_deitsmall8_pretrain.pth
+      checkpoint_file: ${....host.model_root}/backbones/dino_deitsmall8_pretrain.pth
       pretrained: true
     patch_size: 8
     patch_feature_size: 28
@@ -224,28 +139,114 @@ experiment:
       patch_size: null
       patch_overlap: 200
       seg_threshold: 0.5
+  polygonization:
+    method:
+    - acm
+    common_params:
+      init_data_level: 0.5
+    simple_method:
+      data_level: 0.5
+      tolerance:
+      - 1.0
+      seg_threshold: 0.5
+      min_area: 10
+    asm_method:
+      init_method: skeleton
+      data_level: 0.5
+      loss_params:
+        coefs:
+          step_thresholds:
+          - 0
+          - 100
+          - 200
+          - 300
+          data:
+          - 1.0
+          - 0.1
+          - 0.0
+          - 0.0
+          crossfield:
+          - 0.0
+          - 0.05
+          - 0.0
+          - 0.0
+          length:
+          - 0.1
+          - 0.01
+          - 0.0
+          - 0.0
+          curvature:
+          - 0.0
+          - 0.0
+          - 1.0
+          - 0.0
+          corner:
+          - 0.0
+          - 0.0
+          - 0.5
+          - 0.0
+          junction:
+          - 0.0
+          - 0.0
+          - 0.5
+          - 0.0
+        curvature_dissimilarity_threshold: 2
+        corner_angles:
+        - 45
+        - 90
+        - 135
+        corner_angle_threshold: 22.5
+        junction_angles:
+        - 0
+        - 45
+        - 90
+        - 135
+        junction_angle_weights:
+        - 1
+        - 0.01
+        - 0.1
+        - 0.01
+        junction_angle_threshold: 22.5
+      lr: 0.1
+      gamma: 0.995
+      device: cuda
+      tolerance:
+      - 1
+      seg_threshold: 0.5
+      min_area: 10
+    acm_method:
+      steps: 500
+      data_level: 0.5
+      data_coef: 0.1
+      length_coef: 0.4
+      crossfield_coef: 0.5
+      poly_lr: 0.01
+      warmup_iters: 100
+      warmup_factor: 0.1
+      device: cuda
+      tolerance:
+      - 1
+      seg_threshold: 0.5
+      min_area: 10
   name: v0_all_bs4x16
   group_name: v2_${.model.name}
-output_dir: ${.host.data_root}/${.experiment.model.name}_outputs/${.dataset.name}/${.experiment.encoder.in_size}/${.experiment.name}
-checkpoint: latest
-checkpoint_file: null
-save_best: true
-save_latest: true
-save_every: 10
-val_every: 1
-best_val_loss: 10000000.0
-best_val_iou: 0.0
-multi_gpu: true
-device: cuda
-log_to_wandb: true
-num_workers: ${.run_type.num_workers}
-update_pbar_every: ${.host.update_pbar_every}
-country: all
-use_lidar: ${.experiment.encoder.use_lidar}
-use_images: ${.experiment.encoder.use_images}
-eval:
+  country: all
+training:
+  save_best: true
+  save_latest: true
+  save_every: 10
+  val_every: 1
+  best_val_loss: 10000000.0
+  best_val_iou: 0.0
+evaluation:
   split: val
-  pred_file: ${..output_dir}/predictions_${..country}_${.split}/${..checkpoint}.json
+  pred_file: ${..output_dir}/predictions_${..experiment.country}_${.split}/${..checkpoint}.json
   modes:
   - iou
   eval_file: results/metrics
+experiment.name: debug
+output_dir: ${.host.model_root}/${.experiment.model.name}/${.experiment.encoder.in_size}/${.experiment.name}
+checkpoint: best_val_iou
+num_workers: ${.run_type.num_workers}
+image_file: demo_data/image0_CH_val.tif
+lidar_file: demo_data/lidar0_CH_val.copc.laz

ffl/224/v0_all_bs4x16/.hydra/hydra.yaml

@@ -112,18 +112,16 @@ hydra:
     hydra:
     - hydra.mode=RUN
     task:
-    - log_to_wandb=true
-    - host=jz
-    - run_type=release
-    - multi_gpu=true
-    - checkpoint=latest
+    - run_type=debug
+    - host=gin
+    - checkpoint=best_val_iou
     - experiment=ffl_fusion
-    - experiment.name=v0_all_bs4x16
-    - country=all
+    - +image_file=demo_data/image0_CH_val.tif
+    - +lidar_file=demo_data/lidar0_CH_val.copc.laz
   job:
-    name: train
+    name: predict_demo
     chdir: null
-    override_dirname: checkpoint=latest,country=all,experiment.name=v0_all_bs4x16,experiment=ffl_fusion,host=jz,log_to_wandb=true,multi_gpu=true,run_type=release
+    override_dirname: +image_file=demo_data/image0_CH_val.tif,+lidar_file=demo_data/lidar0_CH_val.copc.laz,checkpoint=best_val_iou,experiment=ffl_fusion,host=gin,run_type=debug
     id: ???
     num: ???
     config_name: config
@@ -137,26 +135,28 @@ hydra:
   runtime:
     version: 1.3.2
     version_base: '1.3'
-    cwd: /lustre/fswork/projects/rech/cso/uku93eu/python/PixelsPointsPolygons
+    cwd: /run/netsop/u/home-sam/home/rsulzer/remote_python/pixelspointspolygons
     config_sources:
     - path: hydra.conf
       schema: pkg
       provider: hydra
-    - path: /lustre/fswork/projects/rech/cso/uku93eu/python/PixelsPointsPolygons/config
+    - path: /run/netsop/u/home-sam/home/rsulzer/remote_python/pixelspointspolygons/config
       schema: file
       provider: main
     - path: ''
       schema: structured
       provider: schema
-    output_dir: /lustre/fswork/projects/rech/cso/uku93eu/data/ffl_outputs/lidarpoly/224/v0_all_bs4x16
+    output_dir: /data/rsulzer/PixelsPointsPolygons_output/ffl/224/v0_all_bs4x16
     choices:
+      evaluation: val
+      training: default
       experiment: ffl_fusion
+      [email protected]: asm_acm
       [email protected]: ffl
       [email protected]: early_fusion_vit_cnn
-      dataset: lidarpoly
-      polygonization: asm_acm
-      run_type: release
-      host: jz
+      dataset: p3
+      run_type: debug
+      host: gin
       hydra/env: default
       hydra/callbacks: null
       hydra/job_logging: default

ffl/224/v0_all_bs4x16/.hydra/overrides.yaml

@@ -1,8 +1,6 @@
-- log_to_wandb=true
-- host=jz
-- run_type=release
-- multi_gpu=true
-- checkpoint=latest
+- run_type=debug
+- host=gin
+- checkpoint=best_val_iou
 - experiment=ffl_fusion
-- experiment.name=v0_all_bs4x16
-- country=all
+- +image_file=demo_data/image0_CH_val.tif
+- +lidar_file=demo_data/lidar0_CH_val.copc.laz

ffl/224/v4_image_bs4x16/.hydra/config.yaml

@@ -1,149 +1,55 @@
 host:
   name: gin
-  data_root: /data/rsulzer
+  data_root: /data/rsulzer/${..dataset.name}
+  model_root: /data/rsulzer/${..dataset.name}_output
+  multi_gpu: false
+  device: cuda
   update_pbar_every: 1
+  ldof_exe: /user/rsulzer/home/cpp/line-DOF-metric/build/calculate_DoF
 run_type:
-  name: release
+  name: debug
   batch_size: 16
-  train_subset: null
-  val_subset: null
-  test_subset: null
-  logging: INFO
-  num_workers: 16
-  log_to_wandb: true
-polygonization:
-  method:
-  - acm
-  common_params:
-    init_data_level: 0.5
-  simple_method:
-    data_level: 0.5
-    tolerance:
-    - 1.0
-    seg_threshold: 0.5
-    min_area: 10
-  asm_method:
-    init_method: skeleton
-    data_level: 0.5
-    loss_params:
-      coefs:
-        step_thresholds:
-        - 0
-        - 100
-        - 200
-        - 300
-        data:
-        - 1.0
-        - 0.1
-        - 0.0
-        - 0.0
-        crossfield:
-        - 0.0
-        - 0.05
-        - 0.0
-        - 0.0
-        length:
-        - 0.1
-        - 0.01
-        - 0.0
-        - 0.0
-        curvature:
-        - 0.0
-        - 0.0
-        - 1.0
-        - 0.0
-        corner:
-        - 0.0
-        - 0.0
-        - 0.5
-        - 0.0
-        junction:
-        - 0.0
-        - 0.0
-        - 0.5
-        - 0.0
-      curvature_dissimilarity_threshold: 2
-      corner_angles:
-      - 45
-      - 90
-      - 135
-      corner_angle_threshold: 22.5
-      junction_angles:
-      - 0
-      - 45
-      - 90
-      - 135
-      junction_angle_weights:
-      - 1
-      - 0.01
-      - 0.1
-      - 0.01
-      junction_angle_threshold: 22.5
-    lr: 0.1
-    gamma: 0.995
-    device: cuda
-    tolerance:
-    - 1
-    seg_threshold: 0.5
-    min_area: 10
-  acm_method:
-    steps: 500
-    data_level: 0.5
-    data_coef: 0.1
-    length_coef: 0.4
-    crossfield_coef: 0.5
-    poly_lr: 0.01
-    warmup_iters: 100
-    warmup_factor: 0.1
-    device: cuda
-    tolerance:
-    - 1
-    seg_threshold: 0.5
-    min_area: 10
+  train_subset: 256
+  val_subset: 32
+  test_subset: 32
+  logging: DEBUG
+  num_workers: 0
+  log_to_wandb: false
 dataset:
-  name: p3
+  name: PixelsPointsPolygons
   size: ${..experiment.encoder.in_size}
-  path: ${host.data_root}/PixelsPointsPolygons/data/${.size}
+  path: ${host.data_root}/data/${.size}
   annotations:
-    train: ${..path}/annotations_${...country}_train.json
-    val: ${..path}/annotations_${...country}_val.json
-    test: ${..path}/annotations_${...country}_test.json
+    train: ${..path}/annotations/annotations_${...experiment.country}_train.json
+    val: ${..path}/annotations/annotations_${...experiment.country}_val.json
+    test: ${..path}/annotations/annotations_${...experiment.country}_test.json
   ffl_stats:
-    train: ${..path}/ffl/train/stats-${...country}.pt
-    val: ${..path}/ffl/val/stats-${...country}.pt
-    test: ${..path}/ffl/test/stats-${...country}.pt
+    train: ${..path}/ffl/train/stats-${...experiment.country}.pt
+    val: ${..path}/ffl/val/stats-${...experiment.country}.pt
+    test: ${..path}/ffl/test/stats-${...experiment.country}.pt
   train_subset: ${..run_type.train_subset}
   val_subset: ${..run_type.val_subset}
   test_subset: ${..run_type.test_subset}
 experiment:
   encoder:
-    name: early_fusion_vit_cnn
+    name: vit_cnn
     use_images: true
-    use_lidar: true
+    use_lidar: false
+    type: vit_small_patch${.patch_size}_${.in_size}.dino
+    checkpoint_file: ${...host.model_root}/backbones/dino_deitsmall8_pretrain.pth
+    pretrained: true
     in_size: 224
     in_height: ${.in_size}
     in_width: ${.in_size}
-    in_voxel_size:
-      x: 8
-      'y': 8
-      z: 100
-    max_num_points_per_voxel: 64
-    max_num_voxels:
-      train: 784
-      test: 784
-    out_feature_size: ${..model.decoder.in_feature_size}
-    out_feature_height: ${.out_feature_size}
-    out_feature_width: ${.out_feature_size}
-    out_feature_dim: ${..model.decoder.in_feature_dim}
-    vit:
-      type: vit_small_patch${..patch_size}_${..in_size}.dino
-      checkpoint_file: ${....host.data_root}/checkpoints/backbones/dino_deitsmall8_pretrain.pth
-      pretrained: true
     patch_size: 8
     patch_feature_size: 28
     patch_feature_height: ${.patch_feature_size}
     patch_feature_width: ${.patch_feature_size}
     patch_feature_dim: 384
+    out_feature_size: ${..model.decoder.in_feature_size}
+    out_feature_height: ${.out_feature_size}
+    out_feature_width: ${.out_feature_size}
+    out_feature_dim: ${..model.decoder.in_feature_dim}
     image_mean:
     - 0.0
     - 0.0
@@ -158,7 +64,6 @@ experiment:
     - ColorJitter
     - GaussNoise
     - Normalize
-    max_points_per_voxel: 64
   model:
     name: ffl
     compute_seg: true
@@ -224,33 +129,113 @@ experiment:
       patch_size: null
       patch_overlap: 200
       seg_threshold: 0.5
+  polygonization:
+    method:
+    - acm
+    common_params:
+      init_data_level: 0.5
+    simple_method:
+      data_level: 0.5
+      tolerance:
+      - 1.0
+      seg_threshold: 0.5
+      min_area: 10
+    asm_method:
+      init_method: skeleton
+      data_level: 0.5
+      loss_params:
+        coefs:
+          step_thresholds:
+          - 0
+          - 100
+          - 200
+          - 300
+          data:
+          - 1.0
+          - 0.1
+          - 0.0
+          - 0.0
+          crossfield:
+          - 0.0
+          - 0.05
+          - 0.0
+          - 0.0
+          length:
+          - 0.1
+          - 0.01
+          - 0.0
+          - 0.0
+          curvature:
+          - 0.0
+          - 0.0
+          - 1.0
+          - 0.0
+          corner:
+          - 0.0
+          - 0.0
+          - 0.5
+          - 0.0
+          junction:
+          - 0.0
+          - 0.0
+          - 0.5
+          - 0.0
+        curvature_dissimilarity_threshold: 2
+        corner_angles:
+        - 45
+        - 90
+        - 135
+        corner_angle_threshold: 22.5
+        junction_angles:
+        - 0
+        - 45
+        - 90
+        - 135
+        junction_angle_weights:
+        - 1
+        - 0.01
+        - 0.1
+        - 0.01
+        junction_angle_threshold: 22.5
+      lr: 0.1
+      gamma: 0.995
+      device: cuda
+      tolerance:
+      - 1
+      seg_threshold: 0.5
+      min_area: 10
+    acm_method:
+      steps: 500
+      data_level: 0.5
+      data_coef: 0.1
+      length_coef: 0.4
+      crossfield_coef: 0.5
+      poly_lr: 0.01
+      warmup_iters: 100
+      warmup_factor: 0.1
+      device: cuda
+      tolerance:
+      - 1
+      seg_threshold: 0.5
+      min_area: 10
   name: v4_image_bs4x16
   group_name: v2_${.model.name}
-output_dir: ${.host.data_root}/${.experiment.model.name}_outputs/${.dataset.name}/${.experiment.encoder.in_size}/${.experiment.name}
-checkpoint: best_val_iou
-checkpoint_file: null
-save_best: true
-save_latest: true
-save_every: 10
-val_every: 1
-best_val_loss: 10000000.0
-best_val_iou: 0.0
-multi_gpu: false
-device: cuda
-log_to_wandb: false
-num_workers: ${.run_type.num_workers}
-update_pbar_every: ${.host.update_pbar_every}
-country: Switzerland
-use_lidar: ${.experiment.encoder.use_lidar}
-use_images: ${.experiment.encoder.use_images}
-eval:
+  country: CH
+training:
+  save_best: true
+  save_latest: true
+  save_every: 10
+  val_every: 1
+  best_val_loss: 10000000.0
+  best_val_iou: 0.0
+evaluation:
   split: val
-  pred_file: ${..output_dir}/predictions_${..country}_${.split}/${..checkpoint}.json
+  pred_file: ${..output_dir}/predictions_${..experiment.country}_${.split}/${..checkpoint}.json
   modes:
-  - ldof
-  - coco
   - iou
-  - polis
-  - mta
   eval_file: results/metrics
-  ldof_exe: /user/rsulzer/home/cpp/line-DOF-metric/build/calculate_DoF
+experiment.name: debug
+output_dir: ${.host.model_root}/${.experiment.model.name}/${.experiment.encoder.in_size}/${.experiment.name}
+checkpoint: best_val_iou
+num_workers: ${.run_type.num_workers}
+lidar_file: demo_data/lidar0_CH_val.copc.laz

ffl/224/v4_image_bs4x16/.hydra/hydra.yaml

@@ -112,19 +112,15 @@ hydra:
     hydra:
     - hydra.mode=RUN
     task:
-    - run_type=release
+    - run_type=debug
     - host=gin
-    - device=cuda
-    - log_to_wandb=false
-    - multi_gpu=false
     - checkpoint=best_val_iou
-    - country=Switzerland
-    - eval.split=val
-    - experiment.name=v4_image_bs4x16
+    - experiment=ffl_image
+    - +lidar_file=demo_data/lidar0_CH_val.copc.laz
   job:
-    name: predict
+    name: predict_demo
     chdir: null
-    override_dirname: checkpoint=best_val_iou,country=Switzerland,device=cuda,eval.split=val,experiment.name=v4_image_bs4x16,host=gin,log_to_wandb=false,multi_gpu=false,run_type=release
+    override_dirname: +lidar_file=demo_data/lidar0_CH_val.copc.laz,checkpoint=best_val_iou,experiment=ffl_image,host=gin,run_type=debug
     id: ???
     num: ???
     config_name: config
@@ -149,14 +145,16 @@ hydra:
     - path: ''
       schema: structured
       provider: schema
-    output_dir: /data/rsulzer/ffl_outputs/p3/224/v4_image_bs4x16
+    output_dir: /data/rsulzer/PixelsPointsPolygons_output/ffl/224/v4_image_bs4x16
     choices:
-      experiment: ffl_fusion
+      evaluation: val
+      training: default
+      experiment: ffl_image
+      [email protected]: asm_acm
       [email protected]: ffl
-      [email protected]: early_fusion_vit_cnn
+      [email protected]: vit_cnn
       dataset: p3
-      polygonization: asm_acm
-      run_type: release
+      run_type: debug
       host: gin
       hydra/env: default
       hydra/callbacks: null

ffl/224/v4_image_bs4x16/.hydra/overrides.yaml

@@ -1,9 +1,5 @@
-- run_type=release
+- run_type=debug
 - host=gin
-- device=cuda
-- log_to_wandb=false
-- multi_gpu=false
 - checkpoint=best_val_iou
-- country=Switzerland
-- eval.split=val
-- experiment.name=v4_image_bs4x16
+- experiment=ffl_image
+- +lidar_file=demo_data/lidar0_CH_val.copc.laz

ffl/224/v5_lidar_bs2x16_mnv64/.hydra/config.yaml

@@ -1,118 +1,35 @@
 host:
-  name: g5k
-  data_root: /home/rsulzer/data
-  update_pbar_every: 60
+  name: gin
+  data_root: /data/rsulzer/${..dataset.name}
+  model_root: /data/rsulzer/${..dataset.name}_output
+  multi_gpu: false
+  device: cuda
+  update_pbar_every: 1
+  ldof_exe: /user/rsulzer/home/cpp/line-DOF-metric/build/calculate_DoF
 run_type:
-  name: release
+  name: debug
   batch_size: 16
-  train_subset: null
-  val_subset: null
-  test_subset: null
-  logging: INFO
-  num_workers: 16
-  log_to_wandb: true
-polygonization:
-  method:
-  - acm
-  common_params:
-    init_data_level: 0.5
-  simple_method:
-    data_level: 0.5
-    tolerance:
-    - 1.0
-    seg_threshold: 0.5
-    min_area: 10
-  asm_method:
-    init_method: skeleton
-    data_level: 0.5
-    loss_params:
-      coefs:
-        step_thresholds:
-        - 0
-        - 100
-        - 200
-        - 300
-        data:
-        - 1.0
-        - 0.1
-        - 0.0
-        - 0.0
-        crossfield:
-        - 0.0
-        - 0.05
-        - 0.0
-        - 0.0
-        length:
-        - 0.1
-        - 0.01
-        - 0.0
-        - 0.0
-        curvature:
-        - 0.0
-        - 0.0
-        - 1.0
-        - 0.0
-        corner:
-        - 0.0
-        - 0.0
-        - 0.5
-        - 0.0
-        junction:
-        - 0.0
-        - 0.0
-        - 0.5
-        - 0.0
-      curvature_dissimilarity_threshold: 2
-      corner_angles:
-      - 45
-      - 90
-      - 135
-      corner_angle_threshold: 22.5
-      junction_angles:
-      - 0
-      - 45
-      - 90
-      - 135
-      junction_angle_weights:
-      - 1
-      - 0.01
-      - 0.1
-      - 0.01
-      junction_angle_threshold: 22.5
-    lr: 0.1
-    gamma: 0.995
-    device: cuda
-    tolerance:
-    - 1
-    seg_threshold: 0.5
-    min_area: 10
-  acm_method:
-    steps: 500
-    data_level: 0.5
-    data_coef: 0.1
-    length_coef: 0.4
-    crossfield_coef: 0.5
-    poly_lr: 0.01
-    warmup_iters: 100
-    warmup_factor: 0.1
-    device: cuda
-    tolerance:
-    - 1
-    seg_threshold: 0.5
-    min_area: 10
+  train_subset: 256
+  val_subset: 32
+  test_subset: 32
+  logging: DEBUG
+  num_workers: 0
+  log_to_wandb: false
 dataset:
-  name: lidarpoly
+  name: PixelsPointsPolygons
   size: ${..experiment.encoder.in_size}
-  path: ${host.data_root}/${.name}/${.size}
+  path: ${host.data_root}/data/${.size}
   annotations:
-    train: ${..path}/annotations_${...country}_train.json
-    val: ${..path}/annotations_${...country}_val.json
-    test: ${..path}/annotations_${...country}_test.json
+    train: ${..path}/annotations/annotations_${...experiment.country}_train.json
+    val: ${..path}/annotations/annotations_${...experiment.country}_val.json
+    test: ${..path}/annotations/annotations_${...experiment.country}_test.json
+  ffl_stats:
+    train: ${..path}/ffl/train/stats-${...experiment.country}.pt
+    val: ${..path}/ffl/val/stats-${...experiment.country}.pt
+    test: ${..path}/ffl/test/stats-${...experiment.country}.pt
   train_subset: ${..run_type.train_subset}
   val_subset: ${..run_type.val_subset}
   test_subset: ${..run_type.test_subset}
-  augmentations:
-  - D4
 experiment:
   encoder:
     name: pointpillars_vit_cnn
@@ -135,7 +52,7 @@ experiment:
     out_feature_dim: ${..model.decoder.in_feature_dim}
     vit:
       type: vit_small_patch${..patch_size}_${..in_size}.dino
-      checkpoint_file: ${....host.data_root}/checkpoints/backbones/dino_deitsmall8_pretrain.pth
+      checkpoint_file: ${....host.model_root}/backbones/dino_deitsmall8_pretrain.pth
       pretrained: true
     patch_size: 8
     patch_feature_size: 28
@@ -144,6 +61,7 @@ experiment:
     patch_feature_dim: 384
     augmentations:
     - D4
+    max_points_per_voxel: 64
   model:
     name: ffl
     compute_seg: true
@@ -209,28 +127,113 @@ experiment:
       patch_size: null
       patch_overlap: 200
       seg_threshold: 0.5
+  polygonization:
+    method:
+    - acm
+    common_params:
+      init_data_level: 0.5
+    simple_method:
+      data_level: 0.5
+      tolerance:
+      - 1.0
+      seg_threshold: 0.5
+      min_area: 10
+    asm_method:
+      init_method: skeleton
+      data_level: 0.5
+      loss_params:
+        coefs:
+          step_thresholds:
+          - 0
+          - 100
+          - 200
+          - 300
+          data:
+          - 1.0
+          - 0.1
+          - 0.0
+          - 0.0
+          crossfield:
+          - 0.0
+          - 0.05
+          - 0.0
+          - 0.0
+          length:
+          - 0.1
+          - 0.01
+          - 0.0
+          - 0.0
+          curvature:
+          - 0.0
+          - 0.0
+          - 1.0
+          - 0.0
+          corner:
+          - 0.0
+          - 0.0
+          - 0.5
+          - 0.0
+          junction:
+          - 0.0
+          - 0.0
+          - 0.5
+          - 0.0
+        curvature_dissimilarity_threshold: 2
+        corner_angles:
+        - 45
+        - 90
+        - 135
+        corner_angle_threshold: 22.5
+        junction_angles:
+        - 0
+        - 45
+        - 90
+        - 135
+        junction_angle_weights:
+        - 1
+        - 0.01
+        - 0.1
+        - 0.01
+        junction_angle_threshold: 22.5
+      lr: 0.1
+      gamma: 0.995
+      device: cuda
+      tolerance:
+      - 1
+      seg_threshold: 0.5
+      min_area: 10
+    acm_method:
+      steps: 500
+      data_level: 0.5
+      data_coef: 0.1
+      length_coef: 0.4
+      crossfield_coef: 0.5
+      poly_lr: 0.01
+      warmup_iters: 100
+      warmup_factor: 0.1
+      device: cuda
+      tolerance:
+      - 1
+      seg_threshold: 0.5
+      min_area: 10
   name: v5_lidar_bs2x16_mnv64
   group_name: v2_${.model.name}
-output_dir: ${.host.data_root}/${.experiment.model.name}_outputs/${.dataset.name}/${.experiment.encoder.in_size}/${.experiment.name}
-checkpoint: null
-checkpoint_file: null
-save_best: true
-save_latest: true
-save_every: 10
-val_every: 1
-best_val_loss: 10000000.0
-best_val_iou: 0.0
-multi_gpu: true
-device: cuda
-log_to_wandb: true
-num_workers: ${.run_type.num_workers}
-update_pbar_every: ${.host.update_pbar_every}
-country: Switzerland
-use_lidar: ${.experiment.encoder.use_lidar}
-use_images: ${.experiment.encoder.use_images}
-eval:
+  country: CH
+training:
+  save_best: true
+  save_latest: true
+  save_every: 10
+  val_every: 1
+  best_val_loss: 10000000.0
+  best_val_iou: 0.0
+evaluation:
   split: val
-  pred_file: ${..output_dir}/predictions/${..checkpoint}.json
+  pred_file: ${..output_dir}/predictions_${..experiment.country}_${.split}/${..checkpoint}.json
   modes:
   - iou
   eval_file: results/metrics
+experiment.name: debug
+output_dir: ${.host.model_root}/${.experiment.model.name}/${.experiment.encoder.in_size}/${.experiment.name}
+checkpoint: best_val_iou
+num_workers: ${.run_type.num_workers}
+lidar_file: demo_data/lidar0_CH_val.copc.laz

ffl/224/v5_lidar_bs2x16_mnv64/.hydra/hydra.yaml

@@ -112,16 +112,15 @@ hydra:
     hydra:
     - hydra.mode=RUN
     task:
-    - log_to_wandb=true
-    - host=g5k
-    - run_type=release
-    - multi_gpu=true
-    - checkpoint=null
-    - experiment=lidar_density_ablation64
+    - run_type=debug
+    - host=gin
+    - checkpoint=best_val_iou
+    - experiment=ffl_lidar
+    - +lidar_file=demo_data/lidar0_CH_val.copc.laz
   job:
-    name: train
+    name: predict_demo
     chdir: null
-    override_dirname: checkpoint=null,experiment=lidar_density_ablation64,host=g5k,log_to_wandb=true,multi_gpu=true,run_type=release
+    override_dirname: +lidar_file=demo_data/lidar0_CH_val.copc.laz,checkpoint=best_val_iou,experiment=ffl_lidar,host=gin,run_type=debug
     id: ???
     num: ???
     config_name: config
@@ -135,26 +134,28 @@ hydra:
   runtime:
     version: 1.3.2
     version_base: '1.3'
-    cwd: /home/rsulzer/python/PixelsPointsPolygons
+    cwd: /run/netsop/u/home-sam/home/rsulzer/remote_python/pixelspointspolygons
     config_sources:
     - path: hydra.conf
       schema: pkg
       provider: hydra
-    - path: /home/rsulzer/python/PixelsPointsPolygons/config
+    - path: /run/netsop/u/home-sam/home/rsulzer/remote_python/pixelspointspolygons/config
       schema: file
       provider: main
     - path: ''
       schema: structured
       provider: schema
-    output_dir: /home/rsulzer/data/ffl_outputs/lidarpoly/224/v5_lidar_bs2x16_mnv64
+    output_dir: /data/rsulzer/PixelsPointsPolygons_output/ffl/224/v5_lidar_bs2x16_mnv64
     choices:
-      experiment: lidar_density_ablation64
+      evaluation: val
+      training: default
+      experiment: ffl_lidar
+      [email protected]: asm_acm
       [email protected]: ffl
       [email protected]: pointpillars_vit_cnn
-      dataset: lidarpoly
-      polygonization: asm_acm
-      run_type: release
-      host: g5k
+      dataset: p3
+      run_type: debug
+      host: gin
       hydra/env: default
       hydra/callbacks: null
       hydra/job_logging: default

ffl/224/v5_lidar_bs2x16_mnv64/.hydra/overrides.yaml

@@ -1,6 +1,5 @@
-- log_to_wandb=true
-- host=g5k
-- run_type=release
-- multi_gpu=true
-- checkpoint=null
-- experiment=lidar_density_ablation64
+- run_type=debug
+- host=gin
+- checkpoint=best_val_iou
+- experiment=ffl_lidar
+- +lidar_file=demo_data/lidar0_CH_val.copc.laz

hisup/224/early_fusion_vit_cnn_bs2x16_mnv64/.hydra/config.yaml

@@ -1,117 +1,32 @@
 host:
   name: gin
-  data_root: /data/rsulzer
+  data_root: /data/rsulzer/${..dataset.name}
+  model_root: /data/rsulzer/${..dataset.name}_output
+  multi_gpu: false
+  device: cuda
   update_pbar_every: 1
+  ldof_exe: /user/rsulzer/home/cpp/line-DOF-metric/build/calculate_DoF
 run_type:
-  name: release
+  name: debug
   batch_size: 16
-  train_subset: null
-  val_subset: null
-  test_subset: null
-  logging: INFO
-  num_workers: 16
-  log_to_wandb: true
-polygonization:
-  method:
-  - acm
-  common_params:
-    init_data_level: 0.5
-  simple_method:
-    data_level: 0.5
-    tolerance:
-    - 1.0
-    seg_threshold: 0.5
-    min_area: 10
-  asm_method:
-    init_method: skeleton
-    data_level: 0.5
-    loss_params:
-      coefs:
-        step_thresholds:
-        - 0
-        - 100
-        - 200
-        - 300
-        data:
-        - 1.0
-        - 0.1
-        - 0.0
-        - 0.0
-        crossfield:
-        - 0.0
-        - 0.05
-        - 0.0
-        - 0.0
-        length:
-        - 0.1
-        - 0.01
-        - 0.0
-        - 0.0
-        curvature:
-        - 0.0
-        - 0.0
-        - 1.0
-        - 0.0
-        corner:
-        - 0.0
-        - 0.0
-        - 0.5
-        - 0.0
-        junction:
-        - 0.0
-        - 0.0
-        - 0.5
-        - 0.0
-      curvature_dissimilarity_threshold: 2
-      corner_angles:
-      - 45
-      - 90
-      - 135
-      corner_angle_threshold: 22.5
-      junction_angles:
-      - 0
-      - 45
-      - 90
-      - 135
-      junction_angle_weights:
-      - 1
-      - 0.01
-      - 0.1
-      - 0.01
-      junction_angle_threshold: 22.5
-    lr: 0.1
-    gamma: 0.995
-    device: cuda
-    tolerance:
-    - 1
-    seg_threshold: 0.5
-    min_area: 10
-  acm_method:
-    steps: 500
-    data_level: 0.5
-    data_coef: 0.1
-    length_coef: 0.4
-    crossfield_coef: 0.5
-    poly_lr: 0.01
-    warmup_iters: 100
-    warmup_factor: 0.1
-    device: cuda
-    tolerance:
-    - 1
-    seg_threshold: 0.5
-    min_area: 10
+  train_subset: 256
+  val_subset: 32
+  test_subset: 32
+  logging: DEBUG
+  num_workers: 0
+  log_to_wandb: false
 dataset:
-  name: p3
+  name: PixelsPointsPolygons
   size: ${..experiment.encoder.in_size}
-  path: ${host.data_root}/PixelsPointsPolygons/data/${.size}
+  path: ${host.data_root}/data/${.size}
   annotations:
-    train: ${..path}/annotations_${...country}_train.json
-    val: ${..path}/annotations_${...country}_val.json
-    test: ${..path}/annotations_${...country}_test.json
+    train: ${..path}/annotations/annotations_${...experiment.country}_train.json
+    val: ${..path}/annotations/annotations_${...experiment.country}_val.json
+    test: ${..path}/annotations/annotations_${...experiment.country}_test.json
   ffl_stats:
-    train: ${..path}/ffl/train/stats-${...country}.pt
-    val: ${..path}/ffl/val/stats-${...country}.pt
-    test: ${..path}/ffl/test/stats-${...country}.pt
+    train: ${..path}/ffl/train/stats-${...experiment.country}.pt
+    val: ${..path}/ffl/val/stats-${...experiment.country}.pt
+    test: ${..path}/ffl/test/stats-${...experiment.country}.pt
   train_subset: ${..run_type.train_subset}
   val_subset: ${..run_type.val_subset}
   test_subset: ${..run_type.test_subset}
@@ -137,7 +52,7 @@ experiment:
     out_feature_dim: ${..model.decoder.in_feature_dim}
     vit:
       type: vit_small_patch${..patch_size}_${..in_size}.dino
-      checkpoint_file: ${....host.data_root}/checkpoints/backbones/dino_deitsmall8_pretrain.pth
+      checkpoint_file: ${....host.model_root}/backbones/dino_deitsmall8_pretrain.pth
       pretrained: true
     patch_size: 8
     patch_feature_size: 28
@@ -188,31 +103,23 @@ experiment:
       loss_remask: 1.0
   name: early_fusion_vit_cnn_bs2x16_mnv64
   group_name: v2_${.model.name}
-output_dir: ${.host.data_root}/${.experiment.model.name}_outputs/${.dataset.name}/${.experiment.encoder.in_size}/${.experiment.name}
-checkpoint: best_val_iou
-checkpoint_file: null
-save_best: true
-save_latest: true
-save_every: 10
-val_every: 1
-best_val_loss: 10000000.0
-best_val_iou: 0.0
-multi_gpu: false
-device: cuda
-log_to_wandb: false
-num_workers: ${.run_type.num_workers}
-update_pbar_every: ${.host.update_pbar_every}
-country: CH
-use_lidar: ${.experiment.encoder.use_lidar}
-use_images: ${.experiment.encoder.use_images}
-eval:
+  country: all
+training:
+  save_best: true
+  save_latest: true
+  save_every: 10
+  val_every: 1
+  best_val_loss: 10000000.0
+  best_val_iou: 0.0
+evaluation:
   split: val
-  pred_file: ${..output_dir}/predictions_${..country}_${.split}/${..checkpoint}.json
+  pred_file: ${..output_dir}/predictions_${..experiment.country}_${.split}/${..checkpoint}.json
   modes:
-  - ldof
-  - coco
   - iou
-  - polis
-  - mta
   eval_file: results/metrics
-  ldof_exe: /user/rsulzer/home/cpp/line-DOF-metric/build/calculate_DoF
+experiment.name: debug
+output_dir: ${.host.model_root}/${.experiment.model.name}/${.experiment.encoder.in_size}/${.experiment.name}
+checkpoint: best_val_iou
+num_workers: ${.run_type.num_workers}
+image_file: demo_data/image0_CH_val.tif
+lidar_file: demo_data/lidar0_CH_val.copc.laz

hisup/224/early_fusion_vit_cnn_bs2x16_mnv64/.hydra/hydra.yaml

@@ -112,19 +112,16 @@ hydra:
     hydra:
     - hydra.mode=RUN
     task:
-    - run_type=release
+    - run_type=debug
     - host=gin
-    - device=cuda
-    - log_to_wandb=false
-    - multi_gpu=false
     - checkpoint=best_val_iou
-    - eval.split=val
     - experiment=hisup_fusion
-    - experiment.name=early_fusion_vit_cnn_bs2x16_mnv64
+    - +image_file=demo_data/image0_CH_val.tif
+    - +lidar_file=demo_data/lidar0_CH_val.copc.laz
   job:
-    name: predict
+    name: predict_demo
     chdir: null
-    override_dirname: checkpoint=best_val_iou,device=cuda,eval.split=val,experiment.name=early_fusion_vit_cnn_bs2x16_mnv64,experiment=hisup_fusion,host=gin,log_to_wandb=false,multi_gpu=false,run_type=release
+    override_dirname: +image_file=demo_data/image0_CH_val.tif,+lidar_file=demo_data/lidar0_CH_val.copc.laz,checkpoint=best_val_iou,experiment=hisup_fusion,host=gin,run_type=debug
     id: ???
     num: ???
     config_name: config
@@ -149,14 +146,15 @@ hydra:
     - path: ''
       schema: structured
       provider: schema
-    output_dir: /data/rsulzer/hisup_outputs/p3/224/early_fusion_vit_cnn_bs2x16_mnv64
+    output_dir: /data/rsulzer/PixelsPointsPolygons_output/hisup/224/early_fusion_vit_cnn_bs2x16_mnv64
     choices:
+      evaluation: val
+      training: default
       experiment: hisup_fusion
       [email protected]: hisup
       [email protected]: early_fusion_vit_cnn
       dataset: p3
-      polygonization: asm_acm
-      run_type: release
+      run_type: debug
       host: gin
       hydra/env: default
       hydra/callbacks: null

hisup/224/early_fusion_vit_cnn_bs2x16_mnv64/.hydra/overrides.yaml

@@ -1,9 +1,6 @@
-- run_type=release
+- run_type=debug
 - host=gin
-- device=cuda
-- log_to_wandb=false
-- multi_gpu=false
 - checkpoint=best_val_iou
-- eval.split=val
 - experiment=hisup_fusion
-- experiment.name=early_fusion_vit_cnn_bs2x16_mnv64
+- +image_file=demo_data/image0_CH_val.tif
+- +lidar_file=demo_data/lidar0_CH_val.copc.laz

hisup/224/lidar_pp_vit_cnn_bs2x16_mnv64/.hydra/config.yaml

@@ -1,211 +1,112 @@
-encoder:
-  name: pointpillars_vit_cnn
-  use_images: false
-  use_lidar: true
-  in_size: 224
-  in_height: ${.in_size}
-  in_width: ${.in_size}
-  in_voxel_size:
-    x: 8
-    'y': 8
-    z: 100
-  max_num_points_per_voxel: 64
-  max_num_voxels:
-    train: 784
-    test: 784
-  out_feature_size: ${..model.decoder.in_feature_size}
-  out_feature_height: ${.out_feature_size}
-  out_feature_width: ${.out_feature_size}
-  out_feature_dim: ${..model.decoder.in_feature_dim}
-  vit:
-    type: vit_small_patch${..patch_size}_${..in_size}.dino
-    checkpoint_file: ${...host.data_root}/checkpoints/backbones/dino_deitsmall8_pretrain.pth
-    pretrained: true
-  patch_size: 8
-  patch_feature_size: 28
-  patch_feature_height: ${.patch_feature_size}
-  patch_feature_width: ${.patch_feature_size}
-  patch_feature_dim: 384
-  augmentations:
-  - D4
-model:
-  name: hisup
-  decoder:
-    in_feature_size: 224
-    in_feature_width: ${.in_feature_size}
-    in_feature_height: ${.in_feature_size}
-    in_feature_dim: 256
-  point_pillars:
-    out_channels:
-    - 128
-    - 128
-    - 128
-    upsample_strides:
-    - 1
-    - 2
-    - 4
+host:
+  name: gin
+  data_root: /data/rsulzer/${..dataset.name}
+  model_root: /data/rsulzer/${..dataset.name}_output
+  multi_gpu: false
+  device: cuda
+  update_pbar_every: 1
+  ldof_exe: /user/rsulzer/home/cpp/line-DOF-metric/build/calculate_DoF
+run_type:
+  name: debug
   batch_size: 16
-  start_epoch: 0
-  num_epochs: 200
-  milestone: 0
-  learning_rate: 0.0001
-  weight_decay: 0.0001
-  loss_weights:
-    loss_joff: 0.25
-    loss_jloc: 8.0
-    loss_mask: 1.0
-    loss_afm: 0.1
-    loss_remask: 1.0
-polygonization:
-  method:
-  - acm
-  common_params:
-    init_data_level: 0.5
-  simple_method:
-    data_level: 0.5
-    tolerance:
-    - 1.0
-    seg_threshold: 0.5
-    min_area: 10
-  asm_method:
-    init_method: skeleton
-    data_level: 0.5
-    loss_params:
-      coefs:
-        step_thresholds:
-        - 0
-        - 100
-        - 200
-        - 300
-        data:
-        - 1.0
-        - 0.1
-        - 0.0
-        - 0.0
-        crossfield:
-        - 0.0
-        - 0.05
-        - 0.0
-        - 0.0
-        length:
-        - 0.1
-        - 0.01
-        - 0.0
-        - 0.0
-        curvature:
-        - 0.0
-        - 0.0
-        - 1.0
-        - 0.0
-        corner:
-        - 0.0
-        - 0.0
-        - 0.5
-        - 0.0
-        junction:
-        - 0.0
-        - 0.0
-        - 0.5
-        - 0.0
-      curvature_dissimilarity_threshold: 2
-      corner_angles:
-      - 45
-      - 90
-      - 135
-      corner_angle_threshold: 22.5
-      junction_angles:
-      - 0
-      - 45
-      - 90
-      - 135
-      junction_angle_weights:
-      - 1
-      - 0.01
-      - 0.1
-      - 0.01
-      junction_angle_threshold: 22.5
-    lr: 0.1
-    gamma: 0.995
-    device: cuda
-    tolerance:
-    - 1
-    seg_threshold: 0.5
-    min_area: 10
-  acm_method:
-    steps: 500
-    data_level: 0.5
-    data_coef: 0.1
-    length_coef: 0.4
-    crossfield_coef: 0.5
-    poly_lr: 0.01
-    warmup_iters: 100
-    warmup_factor: 0.1
-    device: cuda
-    tolerance:
-    - 1
-    seg_threshold: 0.5
-    min_area: 10
+  train_subset: 256
+  val_subset: 32
+  test_subset: 32
+  logging: DEBUG
+  num_workers: 0
+  log_to_wandb: false
 dataset:
-  name: lidarpoly
-  size: ${..encoder.in_size}
-  path: ${host.data_root}/${.name}/${.size}
+  name: PixelsPointsPolygons
+  size: ${..experiment.encoder.in_size}
+  path: ${host.data_root}/data/${.size}
   annotations:
-    train: ${..path}/annotations_train.json
-    val: ${..path}/annotations_val.json
-    test: ${..path}/annotations_test.json
+    train: ${..path}/annotations/annotations_${...experiment.country}_train.json
+    val: ${..path}/annotations/annotations_${...experiment.country}_val.json
+    test: ${..path}/annotations/annotations_${...experiment.country}_test.json
+  ffl_stats:
+    train: ${..path}/ffl/train/stats-${...experiment.country}.pt
+    val: ${..path}/ffl/val/stats-${...experiment.country}.pt
+    test: ${..path}/ffl/test/stats-${...experiment.country}.pt
   train_subset: ${..run_type.train_subset}
   val_subset: ${..run_type.val_subset}
-  augmentations:
-  - D4
-run_type:
-  name: release
-  batch_size: 16
-  train_subset: null
-  val_subset: null
-  logging: INFO
-  num_workers: 16
-  log_to_wandb: true
-host:
-  name: g5k
-  data_root: /home/rsulzer/data
-  update_pbar_every: 60
-output_dir: ${.host.data_root}/${.model.name}_outputs/${.dataset.name}/${.encoder.in_size}/${.experiment_name}
-checkpoint: null
-checkpoint_file: null
-experiment_name: lidar_pp_vit_cnn_bs2x16_mnv64
-group_name: v2_${.model.name}
-multi_gpu: true
-device: cuda
-log_to_wandb: true
-num_workers: ${.run_type.num_workers}
-update_pbar_every: ${.host.update_pbar_every}
-use_lidar: ${.encoder.use_lidar}
-use_images: ${.encoder.use_images}
-save_best: true
-save_latest: true
-save_every: 10
-val_every: 1
-best_val_loss: 10000000.0
-best_val_iou: 0.0
-eval:
-  gt_file: ${..dataset.annotations.val}
-  pred_file: ${..output_dir}/predictions/${..checkpoint}.json
+  test_subset: ${..run_type.test_subset}
+experiment:
+  encoder:
+    name: pointpillars_vit_cnn
+    use_images: false
+    use_lidar: true
+    in_size: 224
+    in_height: ${.in_size}
+    in_width: ${.in_size}
+    in_voxel_size:
+      x: 8
+      'y': 8
+      z: 100
+    max_num_points_per_voxel: 64
+    max_num_voxels:
+      train: 784
+      test: 784
+    out_feature_size: ${..model.decoder.in_feature_size}
+    out_feature_height: ${.out_feature_size}
+    out_feature_width: ${.out_feature_size}
+    out_feature_dim: ${..model.decoder.in_feature_dim}
+    vit:
+      type: vit_small_patch${..patch_size}_${..in_size}.dino
+      checkpoint_file: ${....host.model_root}/backbones/dino_deitsmall8_pretrain.pth
+      pretrained: true
+    patch_size: 8
+    patch_feature_size: 28
+    patch_feature_height: ${.patch_feature_size}
+    patch_feature_width: ${.patch_feature_size}
+    patch_feature_dim: 384
+    augmentations:
+    - D4
+  model:
+    name: hisup
+    decoder:
+      in_feature_size: 224
+      in_feature_width: ${.in_feature_size}
+      in_feature_height: ${.in_feature_size}
+      in_feature_dim: 256
+    point_pillars:
+      out_channels:
+      - 128
+      - 128
+      - 128
+      upsample_strides:
+      - 1
+      - 2
+      - 4
+    batch_size: ${...run_type.batch_size}
+    start_epoch: 0
+    num_epochs: 200
+    milestone: 0
+    learning_rate: 0.0001
+    weight_decay: 0.0001
+    loss_weights:
+      loss_joff: 0.25
+      loss_jloc: 8.0
+      loss_mask: 1.0
+      loss_afm: 0.1
+      loss_remask: 1.0
+  name: lidar_pp_vit_cnn_bs2x16_mnv64
+  group_name: v2_${.model.name}
+  country: CH
+training:
+  save_best: true
+  save_latest: true
+  save_every: 10
+  val_every: 1
+  best_val_loss: 10000000.0
+  best_val_iou: 0.0
+evaluation:
+  split: val
+  pred_file: ${..output_dir}/predictions_${..experiment.country}_${.split}/${..checkpoint}.json
   modes:
   - iou
-  experiments:
-  - model: ffl
-    experiment_name:
-    - 224/v3_image_vit_cnn_bs4x16
-    - 224/v3_lidar_pp_vit_cnn_bs2x16
-    - 224/fusion_vit_cnn_bs16
-  - model: hisup
-    experiment_name:
-    - 224/v3_image_hrnet224_bs4x16
-    - 224/v3_lidar_pp_vit_cnn_bs2x16
-    - 224/early_fusion_vit_cnn_bs16
-  - model: pix2poly
-    experiment_name:
-    - 224/image_only_bs4x16
-    - 224/lidar_only_bs2x16
-    - 224/fusion_bs2x16
   eval_file: results/metrics
+experiment.name: debug
+output_dir: ${.host.model_root}/${.experiment.model.name}/${.experiment.encoder.in_size}/${.experiment.name}
+checkpoint: best_val_iou
+num_workers: ${.run_type.num_workers}
+lidar_file: demo_data/lidar0_CH_val.copc.laz

hisup/224/lidar_pp_vit_cnn_bs2x16_mnv64/.hydra/hydra.yaml

@@ -112,19 +112,15 @@ hydra:
     hydra:
     - hydra.mode=RUN
     task:
-    - log_to_wandb=true
-    - host=g5k
-    - run_type=release
-    - multi_gpu=true
-    - experiment_name=lidar_pp_vit_cnn_bs2x16_mnv64
-    - checkpoint=null
-    - model.batch_size=16
-    - encoder=pointpillars_vit_cnn
-    - model=hisup
+    - run_type=debug
+    - host=gin
+    - checkpoint=best_val_iou
+    - experiment=hisup_lidar
+    - +lidar_file=demo_data/lidar0_CH_val.copc.laz
   job:
-    name: train
+    name: predict_demo
     chdir: null
-    override_dirname: checkpoint=null,encoder=pointpillars_vit_cnn,experiment_name=lidar_pp_vit_cnn_bs2x16_mnv64,host=g5k,log_to_wandb=true,model.batch_size=16,model=hisup,multi_gpu=true,run_type=release
+    override_dirname: +lidar_file=demo_data/lidar0_CH_val.copc.laz,checkpoint=best_val_iou,experiment=hisup_lidar,host=gin,run_type=debug
     id: ???
     num: ???
     config_name: config
@@ -138,25 +134,27 @@ hydra:
   runtime:
     version: 1.3.2
     version_base: '1.3'
-    cwd: /home/rsulzer/python/PixelsPointsPolygons
+    cwd: /run/netsop/u/home-sam/home/rsulzer/remote_python/pixelspointspolygons
     config_sources:
     - path: hydra.conf
       schema: pkg
       provider: hydra
-    - path: /home/rsulzer/python/PixelsPointsPolygons/config
+    - path: /run/netsop/u/home-sam/home/rsulzer/remote_python/pixelspointspolygons/config
       schema: file
       provider: main
     - path: ''
       schema: structured
       provider: schema
-    output_dir: /home/rsulzer/data/hisup_outputs/lidarpoly/224/lidar_pp_vit_cnn_bs2x16_mnv64
+    output_dir: /data/rsulzer/PixelsPointsPolygons_output/hisup/224/lidar_pp_vit_cnn_bs2x16_mnv64
     choices:
-      host: g5k
-      run_type: release
-      dataset: lidarpoly
-      polygonization: asm_acm
-      model: hisup
-      encoder: pointpillars_vit_cnn
+      evaluation: val
+      training: default
+      experiment: hisup_lidar
+      [email protected]: hisup
+      [email protected]: pointpillars_vit_cnn
+      dataset: p3
+      run_type: debug
+      host: gin
       hydra/env: default
       hydra/callbacks: null
       hydra/job_logging: default

hisup/224/lidar_pp_vit_cnn_bs2x16_mnv64/.hydra/overrides.yaml

@@ -1,9 +1,5 @@
-- log_to_wandb=true
-- host=g5k
-- run_type=release
-- multi_gpu=true
-- experiment_name=lidar_pp_vit_cnn_bs2x16_mnv64
-- checkpoint=null
-- model.batch_size=16
-- encoder=pointpillars_vit_cnn
-- model=hisup
+- run_type=debug
+- host=gin
+- checkpoint=best_val_iou
+- experiment=hisup_lidar
+- +lidar_file=demo_data/lidar0_CH_val.copc.laz

hisup/224/lidar_pp_vit_cnn_bs2x16_mnv64/predict_demo.log

@@ -0,0 +1 @@
+[2025-05-21 19:42:57,731][timm.models._helpers][ERROR] - No checkpoint found at '/data/rsulzer/PixelsPointsPolygons/checkpoints/backbones/dino_deitsmall8_pretrain.pth'

hisup/224/v3_image_vit_cnn_bs4x12/.hydra/config.yaml

@@ -1,214 +1,115 @@
-encoder:
-  name: vit_cnn
-  use_images: true
-  use_lidar: false
-  type: vit_small_patch${.patch_size}_${.in_size}.dino
-  checkpoint_file: ${..host.data_root}/checkpoints/backbones/dino_deitsmall8_pretrain.pth
-  pretrained: true
-  in_size: 224
-  in_height: ${.in_size}
-  in_width: ${.in_size}
-  patch_size: 8
-  patch_feature_size: 28
-  patch_feature_height: ${.patch_feature_size}
-  patch_feature_width: ${.patch_feature_size}
-  patch_feature_dim: 384
-  out_feature_size: ${..model.decoder.in_feature_size}
-  out_feature_height: ${.out_feature_size}
-  out_feature_width: ${.out_feature_size}
-  out_feature_dim: ${..model.decoder.in_feature_dim}
-  image_mean:
-  - 0.0
-  - 0.0
-  - 0.0
-  image_std:
-  - 1.0
-  - 1.0
-  - 1.0
-  image_max_pixel_value: 255.0
-  augmentations:
-  - D4
-  - ColorJitter
-  - GaussNoise
-  - Normalize
-model:
-  name: hisup
-  decoder:
-    in_feature_size: 224
-    in_feature_width: ${.in_feature_size}
-    in_feature_height: ${.in_feature_size}
-    in_feature_dim: 256
-  point_pillars:
-    out_channels:
-    - 128
-    - 128
-    - 128
-    upsample_strides:
-    - 1
-    - 2
-    - 4
-  batch_size: 12
-  start_epoch: 0
-  num_epochs: 200
-  milestone: 0
-  learning_rate: 0.0001
-  weight_decay: 0.0001
-  loss_weights:
-    loss_joff: 0.25
-    loss_jloc: 8.0
-    loss_mask: 1.0
-    loss_afm: 0.1
-    loss_remask: 1.0
-polygonization:
-  method:
-  - acm
-  common_params:
-    init_data_level: 0.5
-  simple_method:
-    data_level: 0.5
-    tolerance:
-    - 1.0
-    seg_threshold: 0.5
-    min_area: 10
-  asm_method:
-    init_method: skeleton
-    data_level: 0.5
-    loss_params:
-      coefs:
-        step_thresholds:
-        - 0
-        - 100
-        - 200
-        - 300
-        data:
-        - 1.0
-        - 0.1
-        - 0.0
-        - 0.0
-        crossfield:
-        - 0.0
-        - 0.05
-        - 0.0
-        - 0.0
-        length:
-        - 0.1
-        - 0.01
-        - 0.0
-        - 0.0
-        curvature:
-        - 0.0
-        - 0.0
-        - 1.0
-        - 0.0
-        corner:
-        - 0.0
-        - 0.0
-        - 0.5
-        - 0.0
-        junction:
-        - 0.0
-        - 0.0
-        - 0.5
-        - 0.0
-      curvature_dissimilarity_threshold: 2
-      corner_angles:
-      - 45
-      - 90
-      - 135
-      corner_angle_threshold: 22.5
-      junction_angles:
-      - 0
-      - 45
-      - 90
-      - 135
-      junction_angle_weights:
-      - 1
-      - 0.01
-      - 0.1
-      - 0.01
-      junction_angle_threshold: 22.5
-    lr: 0.1
-    gamma: 0.995
-    device: cuda
-    tolerance:
-    - 1
-    seg_threshold: 0.5
-    min_area: 10
-  acm_method:
-    steps: 500
-    data_level: 0.5
-    data_coef: 0.1
-    length_coef: 0.4
-    crossfield_coef: 0.5
-    poly_lr: 0.01
-    warmup_iters: 100
-    warmup_factor: 0.1
-    device: cuda
-    tolerance:
-    - 1
-    seg_threshold: 0.5
-    min_area: 10
+host:
+  name: gin
+  data_root: /data/rsulzer/${..dataset.name}
+  model_root: /data/rsulzer/${..dataset.name}_output
+  multi_gpu: false
+  device: cuda
+  update_pbar_every: 1
+  ldof_exe: /user/rsulzer/home/cpp/line-DOF-metric/build/calculate_DoF
+run_type:
+  name: debug
+  batch_size: 16
+  train_subset: 256
+  val_subset: 32
+  test_subset: 32
+  logging: DEBUG
+  num_workers: 0
+  log_to_wandb: false
 dataset:
-  name: lidarpoly
-  size: ${..encoder.in_size}
-  path: ${host.data_root}/${.name}/${.size}
+  name: PixelsPointsPolygons
+  size: ${..experiment.encoder.in_size}
+  path: ${host.data_root}/data/${.size}
   annotations:
-    train: ${..path}/annotations_train.json
-    val: ${..path}/annotations_val.json
-    test: ${..path}/annotations_test.json
+    train: ${..path}/annotations/annotations_${...experiment.country}_train.json
+    val: ${..path}/annotations/annotations_${...experiment.country}_val.json
+    test: ${..path}/annotations/annotations_${...experiment.country}_test.json
+  ffl_stats:
+    train: ${..path}/ffl/train/stats-${...experiment.country}.pt
+    val: ${..path}/ffl/val/stats-${...experiment.country}.pt
+    test: ${..path}/ffl/test/stats-${...experiment.country}.pt
   train_subset: ${..run_type.train_subset}
   val_subset: ${..run_type.val_subset}
-  augmentations:
-  - D4
-run_type:
-  name: release
-  batch_size: 16
-  train_subset: null
-  val_subset: null
-  logging: INFO
-  num_workers: 16
-  log_to_wandb: true
-host:
-  name: jeanzay
-  data_root: /lustre/fswork/projects/rech/cso/uku93eu/data
-  update_pbar_every: 60
-output_dir: ${.host.data_root}/${.model.name}_outputs/${.dataset.name}/${.encoder.in_size}/${.experiment_name}
-checkpoint: latest
-checkpoint_file: null
-experiment_name: v3_image_vit_cnn_bs4x12
-group_name: v2_${.model.name}
-multi_gpu: true
-device: cuda
-log_to_wandb: true
-num_workers: ${.run_type.num_workers}
-update_pbar_every: ${.host.update_pbar_every}
-use_lidar: ${.encoder.use_lidar}
-use_images: ${.encoder.use_images}
-save_best: true
-save_latest: true
-save_every: 10
-val_every: 1
-best_val_loss: 10000000.0
-best_val_iou: 0.0
-eval:
-  gt_file: ${..dataset.annotations.val}
-  pred_file: ${..output_dir}/predictions/${..checkpoint}.json
+  test_subset: ${..run_type.test_subset}
+experiment:
+  encoder:
+    name: vit_cnn
+    use_images: true
+    use_lidar: false
+    type: vit_small_patch${.patch_size}_${.in_size}.dino
+    checkpoint_file: ${...host.model_root}/backbones/dino_deitsmall8_pretrain.pth
+    pretrained: true
+    in_size: 224
+    in_height: ${.in_size}
+    in_width: ${.in_size}
+    patch_size: 8
+    patch_feature_size: 28
+    patch_feature_height: ${.patch_feature_size}
+    patch_feature_width: ${.patch_feature_size}
+    patch_feature_dim: 384
+    out_feature_size: ${..model.decoder.in_feature_size}
+    out_feature_height: ${.out_feature_size}
+    out_feature_width: ${.out_feature_size}
+    out_feature_dim: ${..model.decoder.in_feature_dim}
+    image_mean:
+    - 0.0
+    - 0.0
+    - 0.0
+    image_std:
+    - 1.0
+    - 1.0
+    - 1.0
+    image_max_pixel_value: 255.0
+    augmentations:
+    - D4
+    - ColorJitter
+    - GaussNoise
+    - Normalize
+  model:
+    name: hisup
+    decoder:
+      in_feature_size: 224
+      in_feature_width: ${.in_feature_size}
+      in_feature_height: ${.in_feature_size}
+      in_feature_dim: 256
+    point_pillars:
+      out_channels:
+      - 128
+      - 128
+      - 128
+      upsample_strides:
+      - 1
+      - 2
+      - 4
+    batch_size: ${...run_type.batch_size}
+    start_epoch: 0
+    num_epochs: 200
+    milestone: 0
+    learning_rate: 0.0001
+    weight_decay: 0.0001
+    loss_weights:
+      loss_joff: 0.25
+      loss_jloc: 8.0
+      loss_mask: 1.0
+      loss_afm: 0.1
+      loss_remask: 1.0
+  name: v3_image_vit_cnn_bs4x12
+  group_name: v2_${.model.name}
+  country: CH
+training:
+  save_best: true
+  save_latest: true
+  save_every: 10
+  val_every: 1
+  best_val_loss: 10000000.0
+  best_val_iou: 0.0
+evaluation:
+  split: val
+  pred_file: ${..output_dir}/predictions_${..experiment.country}_${.split}/${..checkpoint}.json
   modes:
   - iou
-  experiments:
-  - model: ffl
-    experiment_name:
-    - 224/v3_image_vit_cnn_bs4x16
-    - 224/v3_lidar_pp_vit_cnn_bs2x16
-    - 224/fusion_vit_cnn_bs16
-  - model: hisup
-    experiment_name:
-    - 224/v3_image_hrnet224_bs4x16
-    - 224/v3_lidar_pp_vit_cnn_bs2x16
-    - 224/early_fusion_vit_cnn_bs16
-  - model: pix2poly
-    experiment_name:
-    - 224/image_only_bs4x16
-    - 224/lidar_only_bs2x16
-    - 224/fusion_bs2x16
   eval_file: results/metrics
+experiment.name: debug
+output_dir: ${.host.model_root}/${.experiment.model.name}/${.experiment.encoder.in_size}/${.experiment.name}
+checkpoint: best_val_iou
+num_workers: ${.run_type.num_workers}
+image_file: demo_data/image0_CH_val.tif

hisup/224/v3_image_vit_cnn_bs4x12/.hydra/hydra.yaml

@@ -112,19 +112,15 @@ hydra:
     hydra:
     - hydra.mode=RUN
     task:
-    - log_to_wandb=true
-    - host=jz
-    - run_type=release
-    - multi_gpu=true
-    - checkpoint=latest
-    - model.batch_size=12
-    - experiment_name=v3_image_vit_cnn_bs4x12
-    - model=hisup
-    - encoder=vit_cnn
+    - run_type=debug
+    - host=gin
+    - checkpoint=best_val_iou
+    - experiment=hisup_image
+    - +image_file=demo_data/image0_CH_val.tif
   job:
-    name: train
+    name: predict_demo
     chdir: null
-    override_dirname: checkpoint=latest,encoder=vit_cnn,experiment_name=v3_image_vit_cnn_bs4x12,host=jz,log_to_wandb=true,model.batch_size=12,model=hisup,multi_gpu=true,run_type=release
+    override_dirname: +image_file=demo_data/image0_CH_val.tif,checkpoint=best_val_iou,experiment=hisup_image,host=gin,run_type=debug
     id: ???
     num: ???
     config_name: config
@@ -138,25 +134,27 @@ hydra:
   runtime:
     version: 1.3.2
     version_base: '1.3'
-    cwd: /lustre/fswork/projects/rech/cso/uku93eu/python/PixelsPointsPolygons
+    cwd: /run/netsop/u/home-sam/home/rsulzer/remote_python/pixelspointspolygons
     config_sources:
     - path: hydra.conf
       schema: pkg
       provider: hydra
-    - path: /lustre/fswork/projects/rech/cso/uku93eu/python/PixelsPointsPolygons/config
+    - path: /run/netsop/u/home-sam/home/rsulzer/remote_python/pixelspointspolygons/config
       schema: file
       provider: main
     - path: ''
       schema: structured
       provider: schema
-    output_dir: /lustre/fswork/projects/rech/cso/uku93eu/data/hisup_outputs/lidarpoly/224/v3_image_vit_cnn_bs4x12
+    output_dir: /data/rsulzer/PixelsPointsPolygons_output/hisup/224/v3_image_vit_cnn_bs4x12
     choices:
-      host: jz
-      run_type: release
-      dataset: lidarpoly
-      polygonization: asm_acm
-      model: hisup
-      encoder: vit_cnn
+      evaluation: val
+      training: default
+      experiment: hisup_image
+      [email protected]: hisup
+      [email protected]: vit_cnn
+      dataset: p3
+      run_type: debug
+      host: gin
       hydra/env: default
       hydra/callbacks: null
       hydra/job_logging: default

hisup/224/v3_image_vit_cnn_bs4x12/.hydra/overrides.yaml

@@ -1,9 +1,5 @@
-- log_to_wandb=true
-- host=jz
-- run_type=release
-- multi_gpu=true
-- checkpoint=latest
-- model.batch_size=12
-- experiment_name=v3_image_vit_cnn_bs4x12
-- model=hisup
-- encoder=vit_cnn
+- run_type=debug
+- host=gin
+- checkpoint=best_val_iou
+- experiment=hisup_image
+- +image_file=demo_data/image0_CH_val.tif

pix2poly/224/early_fusion_bs2x16_mnv64/.hydra/config.yaml

@@ -1,220 +1,122 @@
-encoder:
-  name: early_fusion_vit
-  use_images: true
-  use_lidar: true
-  in_size: 224
-  in_height: ${.in_size}
-  in_width: ${.in_size}
-  in_voxel_size:
-    x: 8
-    'y': 8
-    z: 100
-  max_num_points_per_voxel: 64
-  max_num_voxels:
-    train: 784
-    test: 784
-  out_feature_width: 28
-  out_feature_height: 28
-  vit:
-    type: vit_small_patch${..patch_size}_${..in_size}.dino
-    checkpoint_file: ${...host.data_root}/checkpoints/backbones/dino_deitsmall8_pretrain.pth
-    pretrained: true
-  patch_size: 8
-  patch_feature_size: 28
-  patch_feature_height: ${.patch_feature_size}
-  patch_feature_width: ${.patch_feature_size}
-  patch_feature_dim: 384
-  num_patches: 784
-  out_feature_dim: ${..model.decoder.in_feature_dim}
-  image_mean:
-  - 0.0
-  - 0.0
-  - 0.0
-  image_std:
-  - 1.0
-  - 1.0
-  - 1.0
-  image_max_pixel_value: 255.0
-  augmentations:
-  - D4
-  - ColorJitter
-  - GaussNoise
-  - Normalize
-model:
-  name: pix2poly
-  decoder:
-    in_feature_size: ${...encoder.patch_feature_size}
-    in_feature_width: ${.in_feature_size}
-    in_feature_height: ${.in_feature_size}
-    in_feature_dim: 256
-  tokenizer:
-    num_bins: ${...encoder.in_size}
-    shuffle_tokens: false
-    n_vertices: 192
-    max_len: null
-    pad_idx: null
-    generation_steps: null
-  fusion: patch_concat
-  sinkhorn_iterations: 100
-  label_smoothing: 0.0
-  vertex_loss_weight: 1.0
-  perm_loss_weight: 10.0
+host:
+  name: gin
+  data_root: /data/rsulzer/${..dataset.name}
+  model_root: /data/rsulzer/${..dataset.name}_output
+  multi_gpu: false
+  device: cuda
+  update_pbar_every: 1
+  ldof_exe: /user/rsulzer/home/cpp/line-DOF-metric/build/calculate_DoF
+run_type:
+  name: debug
   batch_size: 16
-  start_epoch: 0
-  num_epochs: 200
-  milestone: 0
-  learning_rate: 0.0003
-  weight_decay: 0.0001
-polygonization:
-  method:
-  - acm
-  common_params:
-    init_data_level: 0.5
-  simple_method:
-    data_level: 0.5
-    tolerance:
-    - 1.0
-    seg_threshold: 0.5
-    min_area: 10
-  asm_method:
-    init_method: skeleton
-    data_level: 0.5
-    loss_params:
-      coefs:
-        step_thresholds:
-        - 0
-        - 100
-        - 200
-        - 300
-        data:
-        - 1.0
-        - 0.1
-        - 0.0
-        - 0.0
-        crossfield:
-        - 0.0
-        - 0.05
-        - 0.0
-        - 0.0
-        length:
-        - 0.1
-        - 0.01
-        - 0.0
-        - 0.0
-        curvature:
-        - 0.0
-        - 0.0
-        - 1.0
-        - 0.0
-        corner:
-        - 0.0
-        - 0.0
-        - 0.5
-        - 0.0
-        junction:
-        - 0.0
-        - 0.0
-        - 0.5
-        - 0.0
-      curvature_dissimilarity_threshold: 2
-      corner_angles:
-      - 45
-      - 90
-      - 135
-      corner_angle_threshold: 22.5
-      junction_angles:
-      - 0
-      - 45
-      - 90
-      - 135
-      junction_angle_weights:
-      - 1
-      - 0.01
-      - 0.1
-      - 0.01
-      junction_angle_threshold: 22.5
-    lr: 0.1
-    gamma: 0.995
-    device: cuda
-    tolerance:
-    - 1
-    seg_threshold: 0.5
-    min_area: 10
-  acm_method:
-    steps: 500
-    data_level: 0.5
-    data_coef: 0.1
-    length_coef: 0.4
-    crossfield_coef: 0.5
-    poly_lr: 0.01
-    warmup_iters: 100
-    warmup_factor: 0.1
-    device: cuda
-    tolerance:
-    - 1
-    seg_threshold: 0.5
-    min_area: 10
+  train_subset: 256
+  val_subset: 32
+  test_subset: 32
+  logging: DEBUG
+  num_workers: 0
+  log_to_wandb: false
 dataset:
-  name: lidarpoly
-  size: ${..encoder.in_size}
-  path: ${host.data_root}/${.name}/${.size}
+  name: PixelsPointsPolygons
+  size: ${..experiment.encoder.in_size}
+  path: ${host.data_root}/data/${.size}
   annotations:
-    train: ${..path}/annotations_train.json
-    val: ${..path}/annotations_val.json
-    test: ${..path}/annotations_test.json
+    train: ${..path}/annotations/annotations_${...experiment.country}_train.json
+    val: ${..path}/annotations/annotations_${...experiment.country}_val.json
+    test: ${..path}/annotations/annotations_${...experiment.country}_test.json
+  ffl_stats:
+    train: ${..path}/ffl/train/stats-${...experiment.country}.pt
+    val: ${..path}/ffl/val/stats-${...experiment.country}.pt
+    test: ${..path}/ffl/test/stats-${...experiment.country}.pt
   train_subset: ${..run_type.train_subset}
   val_subset: ${..run_type.val_subset}
-  augmentations:
-  - D4
-run_type:
-  name: release
-  batch_size: 16
-  train_subset: null
-  val_subset: null
-  logging: INFO
-  num_workers: 16
-  log_to_wandb: true
-host:
-  name: g5k
-  data_root: /home/rsulzer/data
-  update_pbar_every: 60
-output_dir: ${.host.data_root}/${.model.name}_outputs/${.dataset.name}/${.encoder.in_size}/${.experiment_name}
-checkpoint: null
-checkpoint_file: null
-experiment_name: early_fusion_bs2x16_mnv64
-group_name: v2_${.model.name}
-multi_gpu: true
-device: cuda
-log_to_wandb: true
-num_workers: ${.run_type.num_workers}
-update_pbar_every: ${.host.update_pbar_every}
-use_lidar: ${.encoder.use_lidar}
-use_images: ${.encoder.use_images}
-save_best: true
-save_latest: true
-save_every: 10
-val_every: 1
-best_val_loss: 10000000.0
-best_val_iou: 0.0
-eval:
-  gt_file: ${..dataset.annotations.val}
-  pred_file: ${..output_dir}/predictions/${..checkpoint}.json
+  test_subset: ${..run_type.test_subset}
+experiment:
+  encoder:
+    name: early_fusion_vit
+    use_images: true
+    use_lidar: true
+    in_size: 224
+    in_height: ${.in_size}
+    in_width: ${.in_size}
+    in_voxel_size:
+      x: 8
+      'y': 8
+      z: 100
+    max_num_points_per_voxel: 64
+    max_num_voxels:
+      train: 784
+      test: 784
+    out_feature_width: 28
+    out_feature_height: 28
+    vit:
+      type: vit_small_patch${..patch_size}_${..in_size}.dino
+      checkpoint_file: ${....host.model_root}/backbones/dino_deitsmall8_pretrain.pth
+      pretrained: true
+    patch_size: 8
+    patch_feature_size: 28
+    patch_feature_height: ${.patch_feature_size}
+    patch_feature_width: ${.patch_feature_size}
+    patch_feature_dim: 384
+    num_patches: 784
+    out_feature_dim: ${..model.decoder.in_feature_dim}
+    image_mean:
+    - 0.0
+    - 0.0
+    - 0.0
+    image_std:
+    - 1.0
+    - 1.0
+    - 1.0
+    image_max_pixel_value: 255.0
+    augmentations:
+    - D4
+    - ColorJitter
+    - GaussNoise
+    - Normalize
+  model:
+    name: pix2poly
+    decoder:
+      in_feature_size: ${...encoder.patch_feature_size}
+      in_feature_width: ${.in_feature_size}
+      in_feature_height: ${.in_feature_size}
+      in_feature_dim: 256
+    tokenizer:
+      num_bins: ${...encoder.in_size}
+      shuffle_tokens: false
+      n_vertices: 192
+      max_len: null
+      pad_idx: null
+      generation_steps: null
+    fusion: patch_concat
+    sinkhorn_iterations: 100
+    label_smoothing: 0.0
+    vertex_loss_weight: 1.0
+    perm_loss_weight: 10.0
+    batch_size: ${...run_type.batch_size}
+    start_epoch: 0
+    num_epochs: 200
+    milestone: 0
+    learning_rate: 0.0003
+    weight_decay: 0.0001
+  name: early_fusion_bs2x16_mnv64
+  group_name: v2_${.model.name}
+  country: all
+training:
+  save_best: true
+  save_latest: true
+  save_every: 10
+  val_every: 1
+  best_val_loss: 10000000.0
+  best_val_iou: 0.0
+evaluation:
+  split: val
+  pred_file: ${..output_dir}/predictions_${..experiment.country}_${.split}/${..checkpoint}.json
   modes:
   - iou
-  experiments:
-  - model: ffl
-    experiment_name:
-    - 224/v3_image_vit_cnn_bs4x16
-    - 224/v3_lidar_pp_vit_cnn_bs2x16
-    - 224/fusion_vit_cnn_bs16
-  - model: hisup
-    experiment_name:
-    - 224/v3_image_hrnet224_bs4x16
-    - 224/v3_lidar_pp_vit_cnn_bs2x16
-    - 224/early_fusion_vit_cnn_bs16
-  - model: pix2poly
-    experiment_name:
-    - 224/image_only_bs4x16
-    - 224/lidar_only_bs2x16
-    - 224/fusion_bs2x16
   eval_file: results/metrics
+experiment.name: debug
+output_dir: ${.host.model_root}/${.experiment.model.name}/${.experiment.encoder.in_size}/${.experiment.name}
+checkpoint: best_val_iou
+num_workers: ${.run_type.num_workers}
+image_file: demo_data/image0_CH_val.tif
+lidar_file: demo_data/lidar0_CH_val.copc.laz

pix2poly/224/early_fusion_bs2x16_mnv64/.hydra/hydra.yaml

@@ -112,19 +112,16 @@ hydra:
     hydra:
     - hydra.mode=RUN
     task:
-    - log_to_wandb=true
-    - host=g5k
-    - run_type=release
-    - multi_gpu=true
-    - experiment_name=early_fusion_bs2x16_mnv64
-    - checkpoint=null
-    - model.batch_size=16
-    - encoder=early_fusion_vit
-    - model=pix2poly
+    - run_type=debug
+    - host=gin
+    - checkpoint=best_val_iou
+    - experiment=p2p_fusion
+    - +image_file=demo_data/image0_CH_val.tif
+    - +lidar_file=demo_data/lidar0_CH_val.copc.laz
   job:
-    name: train
+    name: predict_demo
     chdir: null
-    override_dirname: checkpoint=null,encoder=early_fusion_vit,experiment_name=early_fusion_bs2x16_mnv64,host=g5k,log_to_wandb=true,model.batch_size=16,model=pix2poly,multi_gpu=true,run_type=release
+    override_dirname: +image_file=demo_data/image0_CH_val.tif,+lidar_file=demo_data/lidar0_CH_val.copc.laz,checkpoint=best_val_iou,experiment=p2p_fusion,host=gin,run_type=debug
     id: ???
     num: ???
     config_name: config
@@ -138,25 +135,27 @@ hydra:
   runtime:
     version: 1.3.2
     version_base: '1.3'
-    cwd: /home/rsulzer/python/PixelsPointsPolygons
+    cwd: /run/netsop/u/home-sam/home/rsulzer/remote_python/pixelspointspolygons
     config_sources:
     - path: hydra.conf
       schema: pkg
       provider: hydra
-    - path: /home/rsulzer/python/PixelsPointsPolygons/config
+    - path: /run/netsop/u/home-sam/home/rsulzer/remote_python/pixelspointspolygons/config
       schema: file
       provider: main
     - path: ''
       schema: structured
       provider: schema
-    output_dir: /home/rsulzer/data/pix2poly_outputs/lidarpoly/224/early_fusion_bs2x16_mnv64
+    output_dir: /data/rsulzer/PixelsPointsPolygons_output/pix2poly/224/early_fusion_bs2x16_mnv64
     choices:
-      host: g5k
-      run_type: release
-      dataset: lidarpoly
-      polygonization: asm_acm
-      model: pix2poly
-      encoder: early_fusion_vit
+      evaluation: val
+      training: default
+      experiment: p2p_fusion
+      [email protected]: pix2poly
+      [email protected]: early_fusion_vit
+      dataset: p3
+      run_type: debug
+      host: gin
       hydra/env: default
       hydra/callbacks: null
       hydra/job_logging: default

pix2poly/224/early_fusion_bs2x16_mnv64/.hydra/overrides.yaml

@@ -1,9 +1,6 @@
-- log_to_wandb=true
-- host=g5k
-- run_type=release
-- multi_gpu=true
-- experiment_name=early_fusion_bs2x16_mnv64
-- checkpoint=null
-- model.batch_size=16
-- encoder=early_fusion_vit
-- model=pix2poly
+- run_type=debug
+- host=gin
+- checkpoint=best_val_iou
+- experiment=p2p_fusion
+- +image_file=demo_data/image0_CH_val.tif
+- +lidar_file=demo_data/lidar0_CH_val.copc.laz

pix2poly/224/lidar_pp_vit_bs2x16_mnv64/.hydra/config.yaml

@@ -1,220 +1,121 @@
-encoder:
-  name: pointpillars_vit
-  use_images: false
-  use_lidar: true
-  in_size: 224
-  in_height: ${.in_size}
-  in_width: ${.in_size}
-  in_voxel_size:
-    x: 8
-    'y': 8
-    z: 100
-  max_num_points_per_voxel: 64
-  max_num_voxels:
-    train: 784
-    test: 784
-  out_feature_width: 28
-  out_feature_height: 28
-  vit:
-    type: vit_small_patch${..patch_size}_${..in_size}.dino
-    checkpoint_file: ${...host.data_root}/checkpoints/backbones/dino_deitsmall8_pretrain.pth
-    pretrained: true
-  patch_size: 8
-  patch_feature_size: 28
-  patch_feature_height: ${.patch_feature_size}
-  patch_feature_width: ${.patch_feature_size}
-  patch_feature_dim: 384
-  num_patches: 784
-  out_feature_dim: ${..model.decoder.in_feature_dim}
-  image_mean:
-  - 0.0
-  - 0.0
-  - 0.0
-  image_std:
-  - 1.0
-  - 1.0
-  - 1.0
-  image_max_pixel_value: 255.0
-  augmentations:
-  - D4
-  - ColorJitter
-  - GaussNoise
-  - Normalize
-model:
-  name: pix2poly
-  decoder:
-    in_feature_size: ${...encoder.patch_feature_size}
-    in_feature_width: ${.in_feature_size}
-    in_feature_height: ${.in_feature_size}
-    in_feature_dim: 256
-  tokenizer:
-    num_bins: ${...encoder.in_size}
-    shuffle_tokens: false
-    n_vertices: 192
-    max_len: null
-    pad_idx: null
-    generation_steps: null
-  fusion: patch_concat
-  sinkhorn_iterations: 100
-  label_smoothing: 0.0
-  vertex_loss_weight: 1.0
-  perm_loss_weight: 10.0
+host:
+  name: gin
+  data_root: /data/rsulzer/${..dataset.name}
+  model_root: /data/rsulzer/${..dataset.name}_output
+  multi_gpu: false
+  device: cuda
+  update_pbar_every: 1
+  ldof_exe: /user/rsulzer/home/cpp/line-DOF-metric/build/calculate_DoF
+run_type:
+  name: debug
   batch_size: 16
-  start_epoch: 0
-  num_epochs: 200
-  milestone: 0
-  learning_rate: 0.0003
-  weight_decay: 0.0001
-polygonization:
-  method:
-  - acm
-  common_params:
-    init_data_level: 0.5
-  simple_method:
-    data_level: 0.5
-    tolerance:
-    - 1.0
-    seg_threshold: 0.5
-    min_area: 10
-  asm_method:
-    init_method: skeleton
-    data_level: 0.5
-    loss_params:
-      coefs:
-        step_thresholds:
-        - 0
-        - 100
-        - 200
-        - 300
-        data:
-        - 1.0
-        - 0.1
-        - 0.0
-        - 0.0
-        crossfield:
-        - 0.0
-        - 0.05
-        - 0.0
-        - 0.0
-        length:
-        - 0.1
-        - 0.01
-        - 0.0
-        - 0.0
-        curvature:
-        - 0.0
-        - 0.0
-        - 1.0
-        - 0.0
-        corner:
-        - 0.0
-        - 0.0
-        - 0.5
-        - 0.0
-        junction:
-        - 0.0
-        - 0.0
-        - 0.5
-        - 0.0
-      curvature_dissimilarity_threshold: 2
-      corner_angles:
-      - 45
-      - 90
-      - 135
-      corner_angle_threshold: 22.5
-      junction_angles:
-      - 0
-      - 45
-      - 90
-      - 135
-      junction_angle_weights:
-      - 1
-      - 0.01
-      - 0.1
-      - 0.01
-      junction_angle_threshold: 22.5
-    lr: 0.1
-    gamma: 0.995
-    device: cuda
-    tolerance:
-    - 1
-    seg_threshold: 0.5
-    min_area: 10
-  acm_method:
-    steps: 500
-    data_level: 0.5
-    data_coef: 0.1
-    length_coef: 0.4
-    crossfield_coef: 0.5
-    poly_lr: 0.01
-    warmup_iters: 100
-    warmup_factor: 0.1
-    device: cuda
-    tolerance:
-    - 1
-    seg_threshold: 0.5
-    min_area: 10
+  train_subset: 256
+  val_subset: 32
+  test_subset: 32
+  logging: DEBUG
+  num_workers: 0
+  log_to_wandb: false
 dataset:
-  name: lidarpoly
-  size: ${..encoder.in_size}
-  path: ${host.data_root}/${.name}/${.size}
+  name: PixelsPointsPolygons
+  size: ${..experiment.encoder.in_size}
+  path: ${host.data_root}/data/${.size}
   annotations:
-    train: ${..path}/annotations_train.json
-    val: ${..path}/annotations_val.json
-    test: ${..path}/annotations_test.json
+    train: ${..path}/annotations/annotations_${...experiment.country}_train.json
+    val: ${..path}/annotations/annotations_${...experiment.country}_val.json
+    test: ${..path}/annotations/annotations_${...experiment.country}_test.json
+  ffl_stats:
+    train: ${..path}/ffl/train/stats-${...experiment.country}.pt
+    val: ${..path}/ffl/val/stats-${...experiment.country}.pt
+    test: ${..path}/ffl/test/stats-${...experiment.country}.pt
   train_subset: ${..run_type.train_subset}
   val_subset: ${..run_type.val_subset}
-  augmentations:
-  - D4
-run_type:
-  name: release
-  batch_size: 16
-  train_subset: null
-  val_subset: null
-  logging: INFO
-  num_workers: 16
-  log_to_wandb: true
-host:
-  name: g5k
-  data_root: /home/rsulzer/data
-  update_pbar_every: 60
-output_dir: ${.host.data_root}/${.model.name}_outputs/${.dataset.name}/${.encoder.in_size}/${.experiment_name}
-checkpoint: null
-checkpoint_file: null
-experiment_name: lidar_pp_vit_bs2x16_mnv64
-group_name: v2_${.model.name}
-multi_gpu: true
-device: cuda
-log_to_wandb: true
-num_workers: ${.run_type.num_workers}
-update_pbar_every: ${.host.update_pbar_every}
-use_lidar: ${.encoder.use_lidar}
-use_images: ${.encoder.use_images}
-save_best: true
-save_latest: true
-save_every: 10
-val_every: 1
-best_val_loss: 10000000.0
-best_val_iou: 0.0
-eval:
-  gt_file: ${..dataset.annotations.val}
-  pred_file: ${..output_dir}/predictions/${..checkpoint}.json
+  test_subset: ${..run_type.test_subset}
+experiment:
+  encoder:
+    name: pointpillars_vit
+    use_images: false
+    use_lidar: true
+    in_size: 224
+    in_height: ${.in_size}
+    in_width: ${.in_size}
+    in_voxel_size:
+      x: 8
+      'y': 8
+      z: 100
+    max_num_points_per_voxel: 64
+    max_num_voxels:
+      train: 784
+      test: 784
+    out_feature_width: 28
+    out_feature_height: 28
+    vit:
+      type: vit_small_patch${..patch_size}_${..in_size}.dino
+      checkpoint_file: ${....host.model_root}/backbones/dino_deitsmall8_pretrain.pth
+      pretrained: true
+    patch_size: 8
+    patch_feature_size: 28
+    patch_feature_height: ${.patch_feature_size}
+    patch_feature_width: ${.patch_feature_size}
+    patch_feature_dim: 384
+    num_patches: 784
+    out_feature_dim: ${..model.decoder.in_feature_dim}
+    image_mean:
+    - 0.0
+    - 0.0
+    - 0.0
+    image_std:
+    - 1.0
+    - 1.0
+    - 1.0
+    image_max_pixel_value: 255.0
+    augmentations:
+    - D4
+    - ColorJitter
+    - GaussNoise
+    - Normalize
+  model:
+    name: pix2poly
+    decoder:
+      in_feature_size: ${...encoder.patch_feature_size}
+      in_feature_width: ${.in_feature_size}
+      in_feature_height: ${.in_feature_size}
+      in_feature_dim: 256
+    tokenizer:
+      num_bins: ${...encoder.in_size}
+      shuffle_tokens: false
+      n_vertices: 192
+      max_len: null
+      pad_idx: null
+      generation_steps: null
+    fusion: patch_concat
+    sinkhorn_iterations: 100
+    label_smoothing: 0.0
+    vertex_loss_weight: 1.0
+    perm_loss_weight: 10.0
+    batch_size: ${...run_type.batch_size}
+    start_epoch: 0
+    num_epochs: 200
+    milestone: 0
+    learning_rate: 0.0003
+    weight_decay: 0.0001
+  name: lidar_pp_vit_bs2x16_mnv64
+  group_name: v2_${.model.name}
+  country: CH
+training:
+  save_best: true
+  save_latest: true
+  save_every: 10
+  val_every: 1
+  best_val_loss: 10000000.0
+  best_val_iou: 0.0
+evaluation:
+  split: val
+  pred_file: ${..output_dir}/predictions_${..experiment.country}_${.split}/${..checkpoint}.json
   modes:
   - iou
-  experiments:
-  - model: ffl
-    experiment_name:
-    - 224/v3_image_vit_cnn_bs4x16
-    - 224/v3_lidar_pp_vit_cnn_bs2x16
-    - 224/fusion_vit_cnn_bs16
-  - model: hisup
-    experiment_name:
-    - 224/v3_image_hrnet224_bs4x16
-    - 224/v3_lidar_pp_vit_cnn_bs2x16
-    - 224/early_fusion_vit_cnn_bs16
-  - model: pix2poly
-    experiment_name:
-    - 224/image_only_bs4x16
-    - 224/lidar_only_bs2x16
-    - 224/fusion_bs2x16
   eval_file: results/metrics
+experiment.name: debug
+output_dir: ${.host.model_root}/${.experiment.model.name}/${.experiment.encoder.in_size}/${.experiment.name}
+checkpoint: best_val_iou
+num_workers: ${.run_type.num_workers}
+lidar_file: demo_data/lidar0_CH_val.copc.laz

pix2poly/224/lidar_pp_vit_bs2x16_mnv64/.hydra/hydra.yaml

@@ -112,20 +112,15 @@ hydra:
     hydra:
     - hydra.mode=RUN
     task:
-    - log_to_wandb=true
-    - host=g5k
-    - run_type=release
-    - multi_gpu=true
-    - experiment_name=lidar_pp_vit_bs2x16_mnv64
-    - checkpoint=null
-    - model.batch_size=16
-    - encoder=pointpillars_vit
-    - model=pix2poly
-    - model.num_epochs=200
+    - run_type=debug
+    - host=gin
+    - checkpoint=best_val_iou
+    - experiment=p2p_lidar
+    - +lidar_file=demo_data/lidar0_CH_val.copc.laz
   job:
-    name: train
+    name: predict_demo
     chdir: null
-    override_dirname: checkpoint=null,encoder=pointpillars_vit,experiment_name=lidar_pp_vit_bs2x16_mnv64,host=g5k,log_to_wandb=true,model.batch_size=16,model.num_epochs=200,model=pix2poly,multi_gpu=true,run_type=release
+    override_dirname: +lidar_file=demo_data/lidar0_CH_val.copc.laz,checkpoint=best_val_iou,experiment=p2p_lidar,host=gin,run_type=debug
     id: ???
     num: ???
     config_name: config
@@ -139,25 +134,27 @@ hydra:
   runtime:
     version: 1.3.2
     version_base: '1.3'
-    cwd: /home/rsulzer/python/PixelsPointsPolygons
+    cwd: /run/netsop/u/home-sam/home/rsulzer/remote_python/pixelspointspolygons
     config_sources:
     - path: hydra.conf
       schema: pkg
       provider: hydra
-    - path: /home/rsulzer/python/PixelsPointsPolygons/config
+    - path: /run/netsop/u/home-sam/home/rsulzer/remote_python/pixelspointspolygons/config
       schema: file
       provider: main
     - path: ''
       schema: structured
       provider: schema
-    output_dir: /home/rsulzer/data/pix2poly_outputs/lidarpoly/224/lidar_pp_vit_bs2x16_mnv64
+    output_dir: /data/rsulzer/PixelsPointsPolygons_output/pix2poly/224/lidar_pp_vit_bs2x16_mnv64
     choices:
-      host: g5k
-      run_type: release
-      dataset: lidarpoly
-      polygonization: asm_acm
-      model: pix2poly
-      encoder: pointpillars_vit
+      evaluation: val
+      training: default
+      experiment: p2p_lidar
+      [email protected]: pix2poly
+      [email protected]: pointpillars_vit
+      dataset: p3
+      run_type: debug
+      host: gin
       hydra/env: default
       hydra/callbacks: null
       hydra/job_logging: default

pix2poly/224/lidar_pp_vit_bs2x16_mnv64/.hydra/overrides.yaml

@@ -1,10 +1,5 @@
-- log_to_wandb=true
-- host=g5k
-- run_type=release
-- multi_gpu=true
-- experiment_name=lidar_pp_vit_bs2x16_mnv64
-- checkpoint=null
-- model.batch_size=16
-- encoder=pointpillars_vit
-- model=pix2poly
-- model.num_epochs=200
+- run_type=debug
+- host=gin
+- checkpoint=best_val_iou
+- experiment=p2p_lidar
+- +lidar_file=demo_data/lidar0_CH_val.copc.laz

pix2poly/224/v0_all_bs4x16/.hydra/config.yaml

@@ -118,3 +118,5 @@ experiment.name: debug
 output_dir: ${.host.model_root}/${.experiment.model.name}/${.experiment.encoder.in_size}/${.experiment.name}
 checkpoint: null
 num_workers: ${.run_type.num_workers}
+image_file: demo_data/image0_CH_val.tif
+lidar_file: demo_data/lidar0_CH_val.copc.laz

pix2poly/224/v0_all_bs4x16/.hydra/hydra.yaml

@@ -114,11 +114,12 @@ hydra:
     task:
     - run_type=debug
     - host=gin
-    - run_type.log_to_wandb=false
+    - +image_file=demo_data/image0_CH_val.tif
+    - +lidar_file=demo_data/lidar0_CH_val.copc.laz
   job:
-    name: train
+    name: predict_demo
     chdir: null
-    override_dirname: host=gin,run_type.log_to_wandb=false,run_type=debug
+    override_dirname: +image_file=demo_data/image0_CH_val.tif,+lidar_file=demo_data/lidar0_CH_val.copc.laz,host=gin,run_type=debug
     id: ???
     num: ???
     config_name: config

pix2poly/224/v0_all_bs4x16/.hydra/overrides.yaml

@@ -1,3 +1,4 @@
 - run_type=debug
 - host=gin
-- run_type.log_to_wandb=false
+- +image_file=demo_data/image0_CH_val.tif
+- +lidar_file=demo_data/lidar0_CH_val.copc.laz

pix2poly/224/v4_image_vit_bs4x16/.hydra/config.yaml

@@ -1,125 +1,42 @@
 host:
-  name: jeanzay
-  data_root: /lustre/fswork/projects/rech/cso/uku93eu/data
-  update_pbar_every: 60
+  name: gin
+  data_root: /data/rsulzer/${..dataset.name}
+  model_root: /data/rsulzer/${..dataset.name}_output
+  multi_gpu: false
+  device: cuda
+  update_pbar_every: 1
+  ldof_exe: /user/rsulzer/home/cpp/line-DOF-metric/build/calculate_DoF
 run_type:
-  name: release
+  name: debug
   batch_size: 16
-  train_subset: null
-  val_subset: null
-  test_subset: null
-  logging: INFO
-  num_workers: 16
-  log_to_wandb: true
-polygonization:
-  method:
-  - acm
-  common_params:
-    init_data_level: 0.5
-  simple_method:
-    data_level: 0.5
-    tolerance:
-    - 1.0
-    seg_threshold: 0.5
-    min_area: 10
-  asm_method:
-    init_method: skeleton
-    data_level: 0.5
-    loss_params:
-      coefs:
-        step_thresholds:
-        - 0
-        - 100
-        - 200
-        - 300
-        data:
-        - 1.0
-        - 0.1
-        - 0.0
-        - 0.0
-        crossfield:
-        - 0.0
-        - 0.05
-        - 0.0
-        - 0.0
-        length:
-        - 0.1
-        - 0.01
-        - 0.0
-        - 0.0
-        curvature:
-        - 0.0
-        - 0.0
-        - 1.0
-        - 0.0
-        corner:
-        - 0.0
-        - 0.0
-        - 0.5
-        - 0.0
-        junction:
-        - 0.0
-        - 0.0
-        - 0.5
-        - 0.0
-      curvature_dissimilarity_threshold: 2
-      corner_angles:
-      - 45
-      - 90
-      - 135
-      corner_angle_threshold: 22.5
-      junction_angles:
-      - 0
-      - 45
-      - 90
-      - 135
-      junction_angle_weights:
-      - 1
-      - 0.01
-      - 0.1
-      - 0.01
-      junction_angle_threshold: 22.5
-    lr: 0.1
-    gamma: 0.995
-    device: cuda
-    tolerance:
-    - 1
-    seg_threshold: 0.5
-    min_area: 10
-  acm_method:
-    steps: 500
-    data_level: 0.5
-    data_coef: 0.1
-    length_coef: 0.4
-    crossfield_coef: 0.5
-    poly_lr: 0.01
-    warmup_iters: 100
-    warmup_factor: 0.1
-    device: cuda
-    tolerance:
-    - 1
-    seg_threshold: 0.5
-    min_area: 10
+  train_subset: 256
+  val_subset: 32
+  test_subset: 32
+  logging: DEBUG
+  num_workers: 0
+  log_to_wandb: false
 dataset:
-  name: lidarpoly
+  name: PixelsPointsPolygons
   size: ${..experiment.encoder.in_size}
-  path: ${host.data_root}/${.name}/${.size}
+  path: ${host.data_root}/data/${.size}
   annotations:
-    train: ${..path}/annotations_${...country}_train.json
-    val: ${..path}/annotations_${...country}_val.json
-    test: ${..path}/annotations_${...country}_test.json
+    train: ${..path}/annotations/annotations_${...experiment.country}_train.json
+    val: ${..path}/annotations/annotations_${...experiment.country}_val.json
+    test: ${..path}/annotations/annotations_${...experiment.country}_test.json
+  ffl_stats:
+    train: ${..path}/ffl/train/stats-${...experiment.country}.pt
+    val: ${..path}/ffl/val/stats-${...experiment.country}.pt
+    test: ${..path}/ffl/test/stats-${...experiment.country}.pt
   train_subset: ${..run_type.train_subset}
   val_subset: ${..run_type.val_subset}
   test_subset: ${..run_type.test_subset}
-  augmentations:
-  - D4
 experiment:
   encoder:
     name: vit
     use_images: true
     use_lidar: false
     type: vit_small_patch${.patch_size}_${.in_size}.dino
-    checkpoint_file: ${...host.data_root}/checkpoints/backbones/dino_deitsmall8_pretrain.pth
+    checkpoint_file: ${...host.model_root}/backbones/dino_deitsmall8_pretrain.pth
     pretrained: true
     in_size: 224
     in_height: ${.in_size}
@@ -172,28 +89,22 @@ experiment:
     weight_decay: 0.0001
   name: v4_image_vit_bs4x16
   group_name: v2_${.model.name}
-output_dir: ${.host.data_root}/${.experiment.model.name}_outputs/${.dataset.name}/${.experiment.encoder.in_size}/${.experiment.name}
-checkpoint: null
-checkpoint_file: null
-multi_gpu: true
-device: cuda
-log_to_wandb: true
-num_workers: ${.run_type.num_workers}
-update_pbar_every: ${.host.update_pbar_every}
-country: Switzerland
-use_lidar: ${.experiment.encoder.use_lidar}
-use_images: ${.experiment.encoder.use_images}
-save_best: true
-save_latest: true
-save_every: 10
-val_every: 1
-best_val_loss: 10000000.0
-best_val_iou: 0.0
-eval:
+  country: CH
+training:
+  save_best: true
+  save_latest: true
+  save_every: 10
+  val_every: 1
+  best_val_loss: 10000000.0
+  best_val_iou: 0.0
+evaluation:
   split: val
-  pred_file: ${..output_dir}/predictions/${..checkpoint}.json
+  pred_file: ${..output_dir}/predictions_${..experiment.country}_${.split}/${..checkpoint}.json
   modes:
   - iou
-  - polis
-  - mta
   eval_file: results/metrics
+experiment.name: debug
+output_dir: ${.host.model_root}/${.experiment.model.name}/${.experiment.encoder.in_size}/${.experiment.name}
+checkpoint: best_val_iou
+num_workers: ${.run_type.num_workers}
+image_file: demo_data/image0_CH_val.tif

pix2poly/224/v4_image_vit_bs4x16/.hydra/hydra.yaml

@@ -112,16 +112,15 @@ hydra:
     hydra:
     - hydra.mode=RUN
     task:
-    - log_to_wandb=true
-    - host=jz
-    - run_type=release
-    - multi_gpu=true
-    - checkpoint=null
+    - run_type=debug
+    - host=gin
+    - checkpoint=best_val_iou
     - experiment=p2p_image
+    - +image_file=demo_data/image0_CH_val.tif
   job:
-    name: train
+    name: predict_demo
     chdir: null
-    override_dirname: checkpoint=null,experiment=p2p_image,host=jz,log_to_wandb=true,multi_gpu=true,run_type=release
+    override_dirname: +image_file=demo_data/image0_CH_val.tif,checkpoint=best_val_iou,experiment=p2p_image,host=gin,run_type=debug
     id: ???
     num: ???
     config_name: config
@@ -135,26 +134,27 @@ hydra:
   runtime:
     version: 1.3.2
     version_base: '1.3'
-    cwd: /lustre/fswork/projects/rech/cso/uku93eu/python/PixelsPointsPolygons
+    cwd: /run/netsop/u/home-sam/home/rsulzer/remote_python/pixelspointspolygons
     config_sources:
     - path: hydra.conf
       schema: pkg
       provider: hydra
-    - path: /lustre/fswork/projects/rech/cso/uku93eu/python/PixelsPointsPolygons/config
+    - path: /run/netsop/u/home-sam/home/rsulzer/remote_python/pixelspointspolygons/config
       schema: file
       provider: main
     - path: ''
       schema: structured
       provider: schema
-    output_dir: /lustre/fswork/projects/rech/cso/uku93eu/data/pix2poly_outputs/lidarpoly/224/v4_image_vit_bs4x16
+    output_dir: /data/rsulzer/PixelsPointsPolygons_output/pix2poly/224/v4_image_vit_bs4x16
     choices:
+      evaluation: val
+      training: default
       experiment: p2p_image
       [email protected]: pix2poly
       [email protected]: vit
-      dataset: lidarpoly
-      polygonization: asm_acm
-      run_type: release
-      host: jz
+      dataset: p3
+      run_type: debug
+      host: gin
       hydra/env: default
       hydra/callbacks: null
       hydra/job_logging: default

pix2poly/224/v4_image_vit_bs4x16/.hydra/overrides.yaml

@@ -1,6 +1,5 @@
-- log_to_wandb=true
-- host=jz
-- run_type=release
-- multi_gpu=true
-- checkpoint=null
+- run_type=debug
+- host=gin
+- checkpoint=best_val_iou
 - experiment=p2p_image
+- +image_file=demo_data/image0_CH_val.tif