README.md

---
library_name: autogluon
tags:
- image-classification
- autogluon
- automl
- sign-identification
datasets:
- ecopus/sign_identification
metrics:
- accuracy
- f1
model-index:
- name: sign-identification-autogluon
  results:
  - task:
      type: image-classification
      name: Image Classification
    dataset:
      type: ecopus/sign_identification
      name: Sign Identification
    metrics:
    - type: accuracy
      value: 0.833
      name: Test Accuracy
    - type: f1
      value: 0.829
      name: Test F1 Score (Weighted)
---

# Sign Identification with AutoGluon

This model was trained using AutoGluon's AutoML capabilities for sign identification.
Performance not good enough because of the limited amount of data (only 30 images).


## Model Description

- **Framework**: AutoGluon MultiModal  
- **Task**: Multi-class image classification (N classes)  
- **Dataset**: [ecopus/sign_identification](https://huggingface.co/datasets/ecopus/sign_identification)  
- **Architecture**: ResNet18 (`timm_image`)  
- **Training**: `medium_quality` preset  

## Performance

| Metric | Validation | Test |
|--------|------------|------|
| Accuracy | 0.833 | 0.571 |
| F1 Score (Weighted) | 0.829 | 0.571 |


## Usage

### Download and Load Model (Recommended — Native Directory)

```python
from autogluon.multimodal import MultiModalPredictor
from huggingface_hub import hf_hub_download
import zipfile
import os

# Download the zipped predictor directory
zip_path = hf_hub_download(
    repo_id="yusenthebot/sign-identification-autogluon",
    filename="autogluon_sign_predictor_dir.zip"
)

# Extract
extract_dir = "predictor_dir"
os.makedirs(extract_dir, exist_ok=True)
with zipfile.ZipFile(zip_path, "r") as zf:
    zf.extractall(extract_dir)

# Load predictor
predictor = MultiModalPredictor.load(extract_dir)

# Predict (replace `your_dataframe` with a pandas DataFrame that matches training schema)
# predictions = predictor.predict(your_dataframe)