--- library_name: pytorch license: other tags: - backbone - bu_auto - android pipeline_tag: image-classification --- ![](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/swin_small/web-assets/model_demo.png) # Swin-Small: Optimized for Qualcomm Devices SwinSmall is a machine learning model that can classify images from the Imagenet dataset. It can also be used as a backbone in building more complex models for specific use cases. This is based on the implementation of Swin-Small found [here](https://github.com/pytorch/vision/blob/main/torchvision/models/swin_transformer.py). This repository contains pre-exported model files optimized for Qualcomm® devices. You can use the [Qualcomm® AI Hub Models](https://github.com/qualcomm/ai-hub-models/blob/main/src/qai_hub_models/models/swin_small) library to export with custom configurations. More details on model performance across various devices, can be found [here](#performance-summary). Qualcomm AI Hub Models uses [Qualcomm AI Hub Workbench](https://workbench.aihub.qualcomm.com) to compile, profile, and evaluate this model. [Sign up](https://myaccount.qualcomm.com/signup) to run these models on a hosted Qualcomm® device. ## Getting Started There are two ways to deploy this model on your device: ### Option 1: Download Pre-Exported Models Below are pre-exported model assets ready for deployment. | Runtime | Precision | Chipset | SDK Versions | Download | |---|---|---|---|---| | ONNX | float | Universal | QAIRT 2.42, ONNX Runtime 1.24.3 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/swin_small/releases/v0.54.0/swin_small-onnx-float.zip) | ONNX | w8a16 | Universal | QAIRT 2.42, ONNX Runtime 1.24.3 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/swin_small/releases/v0.54.0/swin_small-onnx-w8a16.zip) | QNN_DLC | float | Universal | QAIRT 2.45 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/swin_small/releases/v0.54.0/swin_small-qnn_dlc-float.zip) | QNN_DLC | w8a16 | Universal | QAIRT 2.45 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/swin_small/releases/v0.54.0/swin_small-qnn_dlc-w8a16.zip) | TFLITE | float | Universal | QAIRT 2.45 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/swin_small/releases/v0.54.0/swin_small-tflite-float.zip) For more device-specific assets and performance metrics, visit **[Swin-Small on Qualcomm® AI Hub](https://aihub.qualcomm.com/models/swin_small)**. ### Option 2: Export with Custom Configurations Use the [Qualcomm® AI Hub Models](https://github.com/qualcomm/ai-hub-models/blob/main/src/qai_hub_models/models/swin_small) Python library to compile and export the model with your own: - Custom weights (e.g., fine-tuned checkpoints) - Custom input shapes - Target device and runtime configurations This option is ideal if you need to customize the model beyond the default configuration provided here. See our repository for [Swin-Small on GitHub](https://github.com/qualcomm/ai-hub-models/blob/main/src/qai_hub_models/models/swin_small) for usage instructions. ## Model Details **Model Type:** Model_use_case.image_classification **Model Stats:** - Model checkpoint: Imagenet - Input resolution: 224x224 - Number of parameters: 50.4M - Model size (float): 193 MB - Model size (w8a16): 52.5 MB ## Performance Summary | Model | Runtime | Precision | Chipset | Inference Time (ms) | Peak Memory Range (MB) | Primary Compute Unit |---|---|---|---|---|---|--- | Swin-Small | ONNX | float | Snapdragon® 8 Elite Gen 5 Mobile | 6.318 ms | 1 - 647 MB | NPU | Swin-Small | ONNX | float | Snapdragon® 8 Elite Mobile | 7.658 ms | 1 - 645 MB | NPU | Swin-Small | ONNX | float | Snapdragon® X2 Elite | 6.719 ms | 101 - 101 MB | NPU | Swin-Small | ONNX | float | Snapdragon® X Elite | 15.881 ms | 100 - 100 MB | NPU | Swin-Small | ONNX | float | Snapdragon® X Elite | 15.881 ms | 100 - 100 MB | NPU | Swin-Small | ONNX | float | Snapdragon® 8 Gen 3 Mobile | 9.989 ms | 0 - 519 MB | NPU | Swin-Small | ONNX | float | Qualcomm® QCS8550 (Proxy) | 15.343 ms | 1 - 7 MB | NPU | Swin-Small | ONNX | float | Snapdragon® 8 Elite For Galaxy Mobile | 7.658 ms | 1 - 645 MB | NPU | Swin-Small | ONNX | float | Qualcomm® QCS9075 | 19.924 ms | 0 - 4 MB | NPU | Swin-Small | ONNX | w8a16 | Snapdragon® 8 Elite Gen 5 Mobile | 5.887 ms | 0 - 382 MB | NPU | Swin-Small | ONNX | w8a16 | Snapdragon® 8 Elite Mobile | 7.337 ms | 0 - 458 MB | NPU | Swin-Small | ONNX | w8a16 | Snapdragon® X2 Elite | 6.325 ms | 54 - 54 MB | NPU | Swin-Small | ONNX | w8a16 | Snapdragon® X Elite | 14.567 ms | 53 - 53 MB | NPU | Swin-Small | ONNX | w8a16 | Snapdragon® X Elite | 14.567 ms | 53 - 53 MB | NPU | Swin-Small | ONNX | w8a16 | Snapdragon® 8 Gen 3 Mobile | 9.211 ms | 0 - 528 MB | NPU | Swin-Small | ONNX | w8a16 | Qualcomm® QCS6490 | 721.617 ms | 94 - 109 MB | CPU | Swin-Small | ONNX | w8a16 | Qualcomm® QCS8550 (Proxy) | 13.958 ms | 0 - 62 MB | NPU | Swin-Small | ONNX | w8a16 | Snapdragon® 8 Elite For Galaxy Mobile | 7.337 ms | 0 - 458 MB | NPU | Swin-Small | ONNX | w8a16 | Snapdragon® 7 Gen 4 Mobile | 355.219 ms | 92 - 113 MB | CPU | Swin-Small | ONNX | w8a16 | Qualcomm® QCM6690 | 379.333 ms | 75 - 94 MB | CPU | Swin-Small | ONNX | w8a16 | Qualcomm® QCS9075 | 16.989 ms | 0 - 3 MB | NPU | Swin-Small | ONNX | w8a16 | Snapdragon® 7 Gen 4 Mobile | 355.219 ms | 92 - 113 MB | CPU | Swin-Small | QNN_DLC | float | Snapdragon® 8 Elite Gen 5 Mobile | 6.213 ms | 0 - 596 MB | NPU | Swin-Small | QNN_DLC | float | Snapdragon® 8 Elite Mobile | 7.588 ms | 1 - 588 MB | NPU | Swin-Small | QNN_DLC | float | Snapdragon® X2 Elite | 7.001 ms | 1 - 1 MB | NPU | Swin-Small | QNN_DLC | float | Snapdragon® X Elite | 16.225 ms | 1 - 1 MB | NPU | Swin-Small | QNN_DLC | float | Snapdragon® X Elite | 16.225 ms | 1 - 1 MB | NPU | Swin-Small | QNN_DLC | float | Snapdragon® 8 Gen 3 Mobile | 9.956 ms | 1 - 408 MB | NPU | Swin-Small | QNN_DLC | float | Qualcomm® QCS8550 (Proxy) | 15.434 ms | 1 - 2 MB | NPU | Swin-Small | QNN_DLC | float | Qualcomm® SA8775P | 17.415 ms | 1 - 592 MB | NPU | Swin-Small | QNN_DLC | float | Qualcomm® SA8775P | 17.415 ms | 1 - 592 MB | NPU | Swin-Small | QNN_DLC | float | Qualcomm® SA8775P | 17.415 ms | 1 - 592 MB | NPU | Swin-Small | QNN_DLC | float | Qualcomm® SA7255P | 37.931 ms | 1 - 275 MB | NPU | Swin-Small | QNN_DLC | float | Qualcomm® QCS8450 (Proxy) | 23.988 ms | 0 - 395 MB | NPU | Swin-Small | QNN_DLC | float | Qualcomm® SA8295P | 22.638 ms | 1 - 269 MB | NPU | Swin-Small | QNN_DLC | float | Snapdragon® 8 Elite For Galaxy Mobile | 7.588 ms | 1 - 588 MB | NPU | Swin-Small | QNN_DLC | float | Qualcomm® QCS9075 | 19.566 ms | 1 - 3 MB | NPU | Swin-Small | QNN_DLC | w8a16 | Snapdragon® 8 Elite Gen 5 Mobile | 6.396 ms | 0 - 635 MB | NPU | Swin-Small | QNN_DLC | w8a16 | Snapdragon® 8 Elite Mobile | 8.079 ms | 0 - 622 MB | NPU | Swin-Small | QNN_DLC | w8a16 | Snapdragon® X2 Elite | 7.205 ms | 0 - 0 MB | NPU | Swin-Small | QNN_DLC | w8a16 | Snapdragon® X Elite | 17.263 ms | 0 - 0 MB | NPU | Swin-Small | QNN_DLC | w8a16 | Snapdragon® X Elite | 17.263 ms | 0 - 0 MB | NPU | Swin-Small | QNN_DLC | w8a16 | Snapdragon® 8 Gen 3 Mobile | 10.784 ms | 0 - 863 MB | NPU | Swin-Small | QNN_DLC | w8a16 | Qualcomm® QCS8550 (Proxy) | 16.26 ms | 0 - 4 MB | NPU | Swin-Small | QNN_DLC | w8a16 | Qualcomm® SA8775P | 16.656 ms | 0 - 638 MB | NPU | Swin-Small | QNN_DLC | w8a16 | Qualcomm® SA8775P | 16.656 ms | 0 - 638 MB | NPU | Swin-Small | QNN_DLC | w8a16 | Qualcomm® SA8775P | 16.656 ms | 0 - 638 MB | NPU | Swin-Small | QNN_DLC | w8a16 | Qualcomm® SA7255P | 29.084 ms | 0 - 634 MB | NPU | Swin-Small | QNN_DLC | w8a16 | Snapdragon® 8 Elite For Galaxy Mobile | 8.079 ms | 0 - 622 MB | NPU | Swin-Small | QNN_DLC | w8a16 | Snapdragon® 7 Gen 4 Mobile | 16.962 ms | 0 - 654 MB | NPU | Swin-Small | QNN_DLC | w8a16 | Qualcomm® QCM6690 | 73.6 ms | 0 - 625 MB | NPU | Swin-Small | QNN_DLC | w8a16 | Qualcomm® QCS9075 | 19.102 ms | 2 - 4 MB | NPU | Swin-Small | QNN_DLC | w8a16 | Snapdragon® 7 Gen 4 Mobile | 16.962 ms | 0 - 654 MB | NPU | Swin-Small | TFLITE | float | Snapdragon® 8 Elite Gen 5 Mobile | 6.528 ms | 0 - 310 MB | NPU | Swin-Small | TFLITE | float | Snapdragon® 8 Elite Mobile | 7.798 ms | 0 - 284 MB | NPU | Swin-Small | TFLITE | float | Snapdragon® 8 Gen 3 Mobile | 10.168 ms | 0 - 436 MB | NPU | Swin-Small | TFLITE | float | Qualcomm® QCS8550 (Proxy) | 15.622 ms | 0 - 4 MB | NPU | Swin-Small | TFLITE | float | Qualcomm® SA8775P | 17.463 ms | 0 - 305 MB | NPU | Swin-Small | TFLITE | float | Qualcomm® SA8775P | 17.463 ms | 0 - 305 MB | NPU | Swin-Small | TFLITE | float | Qualcomm® SA8775P | 17.463 ms | 0 - 305 MB | NPU | Swin-Small | TFLITE | float | Qualcomm® SA7255P | 37.447 ms | 0 - 301 MB | NPU | Swin-Small | TFLITE | float | Qualcomm® QCS8450 (Proxy) | 23.5 ms | 0 - 410 MB | NPU | Swin-Small | TFLITE | float | Qualcomm® SA8295P | 22.961 ms | 0 - 296 MB | NPU | Swin-Small | TFLITE | float | Snapdragon® 8 Elite For Galaxy Mobile | 7.798 ms | 0 - 284 MB | NPU | Swin-Small | TFLITE | float | Qualcomm® QCS9075 | 19.903 ms | 0 - 104 MB | NPU ## License * The license for the original implementation of Swin-Small can be found [here](https://github.com/pytorch/vision/blob/main/LICENSE). ## References * [Swin Transformer: Hierarchical Vision Transformer using Shifted Windows](https://arxiv.org/abs/2103.14030) * [Source Model Implementation](https://github.com/pytorch/vision/blob/main/torchvision/models/swin_transformer.py) ## Community * Join [our AI Hub Slack community](https://aihub.qualcomm.com/community/slack) to collaborate, post questions and learn more about on-device AI. * For questions or feedback please [reach out to us](mailto:ai-hub-support@qti.qualcomm.com).