README.md

---
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/swinv2_base/web-assets/model_demo.png)

# SwinV2-Base: Optimized for Qualcomm Devices

SwinV2Base 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 SwinV2-Base 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/v0.59.0/src/qai_hub_models/models/swinv2_base) 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 |
|---|---|---|---|---|
| QNN_DLC | float | Universal | QAIRT 2.45 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/swinv2_base/releases/v0.59.0/swinv2_base-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/swinv2_base/releases/v0.59.0/swinv2_base-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/swinv2_base/releases/v0.59.0/swinv2_base-tflite-float.zip)

For more device-specific assets and performance metrics, visit **[SwinV2-Base on Qualcomm® AI Hub](https://aihub.qualcomm.com/models/swinv2_base)**.


### Option 2: Export with Custom Configurations

Use the [Qualcomm® AI Hub Models](https://github.com/qualcomm/ai-hub-models/blob/v0.59.0/src/qai_hub_models/models/swinv2_base) 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 [SwinV2-Base on GitHub](https://github.com/qualcomm/ai-hub-models/blob/v0.59.0/src/qai_hub_models/models/swinv2_base) for usage instructions.

## Model Details

**Model Type:** Model_use_case.image_classification

**Model Stats:**
- Model checkpoint: Imagenet
- Input resolution: 256x256
- Number of parameters: 88.8M
- Model size (float): 339 MB
- Model size (w8a16): 90.2 MB

## Performance Summary
| Model | Runtime | Precision | Chipset | Inference Time (ms) | Peak Memory Range (MB) | Primary Compute Unit
|---|---|---|---|---|---|---
| SwinV2-Base | QNN_DLC | float | Snapdragon® X2 Elite | 12.451 ms | 1 - 1 MB | NPU
| SwinV2-Base | QNN_DLC | float | Snapdragon® X Elite | 28.955 ms | 1 - 1 MB | NPU
| SwinV2-Base | QNN_DLC | float | Snapdragon® 8 Gen 3 Mobile | 19.714 ms | 1 - 545 MB | NPU
| SwinV2-Base | QNN_DLC | float | Snapdragon® 8 Gen 1 Mobile | 41.129 ms | 0 - 534 MB | NPU
| SwinV2-Base | QNN_DLC | float | Qualcomm® Dragonwing™ QCS8275 | 74.071 ms | 1 - 395 MB | NPU
| SwinV2-Base | QNN_DLC | float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 27.948 ms | 1 - 449 MB | NPU
| SwinV2-Base | QNN_DLC | float | Qualcomm® SA8775P | 31.464 ms | 1 - 388 MB | NPU
| SwinV2-Base | QNN_DLC | float | Qualcomm® SA8650P | 31.464 ms | 1 - 388 MB | NPU
| SwinV2-Base | QNN_DLC | float | Qualcomm® SA8255P | 31.464 ms | 1 - 388 MB | NPU
| SwinV2-Base | QNN_DLC | float | Qualcomm® QCS8450 | 41.129 ms | 0 - 534 MB | NPU
| SwinV2-Base | QNN_DLC | float | Qualcomm® Dragonwing™ IQ-9075 | 39.558 ms | 3 - 5 MB | NPU
| SwinV2-Base | QNN_DLC | float | Qualcomm® Dragonwing™ IQ-X7181 | 28.955 ms | 1 - 1 MB | NPU
| SwinV2-Base | QNN_DLC | float | Qualcomm® Dragonwing™ Q-8750 | 14.613 ms | 1 - 392 MB | NPU
| SwinV2-Base | QNN_DLC | float | Qualcomm® SA7255P | 74.071 ms | 1 - 395 MB | NPU
| SwinV2-Base | QNN_DLC | float | Qualcomm® SA8295P | 37.795 ms | 1 - 378 MB | NPU
| SwinV2-Base | QNN_DLC | float | Snapdragon® 8 Elite Mobile | 14.613 ms | 1 - 392 MB | NPU
| SwinV2-Base | QNN_DLC | float | Snapdragon® 8 Elite Gen 5 Mobile | 11.583 ms | 0 - 424 MB | NPU
| SwinV2-Base | QNN_DLC | w8a16 | Snapdragon® X2 Elite | 12.215 ms | 0 - 0 MB | NPU
| SwinV2-Base | QNN_DLC | w8a16 | Snapdragon® X Elite | 30.702 ms | 0 - 0 MB | NPU
| SwinV2-Base | QNN_DLC | w8a16 | Snapdragon® 8 Gen 3 Mobile | 19.654 ms | 0 - 2036 MB | NPU
| SwinV2-Base | QNN_DLC | w8a16 | Qualcomm® Dragonwing™ QCS8275 | 52.667 ms | 0 - 486 MB | NPU
| SwinV2-Base | QNN_DLC | w8a16 | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 29.334 ms | 0 - 3 MB | NPU
| SwinV2-Base | QNN_DLC | w8a16 | Qualcomm® SA8775P | 29.997 ms | 0 - 869 MB | NPU
| SwinV2-Base | QNN_DLC | w8a16 | Qualcomm® SA8650P | 29.997 ms | 0 - 869 MB | NPU
| SwinV2-Base | QNN_DLC | w8a16 | Qualcomm® SA8255P | 29.997 ms | 0 - 869 MB | NPU
| SwinV2-Base | QNN_DLC | w8a16 | Qualcomm® Dragonwing™ IQ-9075 | 29.947 ms | 0 - 2 MB | NPU
| SwinV2-Base | QNN_DLC | w8a16 | Qualcomm® Dragonwing™ IQ-X7181 | 30.702 ms | 0 - 0 MB | NPU
| SwinV2-Base | QNN_DLC | w8a16 | Qualcomm® Dragonwing™ Q-8750 | 14.736 ms | 0 - 903 MB | NPU
| SwinV2-Base | QNN_DLC | w8a16 | Qualcomm® SA7255P | 52.667 ms | 0 - 486 MB | NPU
| SwinV2-Base | QNN_DLC | w8a16 | Snapdragon® 8 Elite Mobile | 14.736 ms | 0 - 903 MB | NPU
| SwinV2-Base | QNN_DLC | w8a16 | Snapdragon® 8 Elite Gen 5 Mobile | 11.415 ms | 0 - 953 MB | NPU
| SwinV2-Base | TFLITE | float | Snapdragon® 8 Gen 3 Mobile | 19.963 ms | 0 - 2174 MB | NPU
| SwinV2-Base | TFLITE | float | Snapdragon® 8 Gen 1 Mobile | 42.639 ms | 0 - 676 MB | NPU
| SwinV2-Base | TFLITE | float | Qualcomm® Dragonwing™ QCS8275 | 71.105 ms | 0 - 885 MB | NPU
| SwinV2-Base | TFLITE | float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 29.556 ms | 0 - 4 MB | NPU
| SwinV2-Base | TFLITE | float | Qualcomm® SA8775P | 32.555 ms | 0 - 879 MB | NPU
| SwinV2-Base | TFLITE | float | Qualcomm® SA8650P | 32.555 ms | 0 - 879 MB | NPU
| SwinV2-Base | TFLITE | float | Qualcomm® SA8255P | 32.555 ms | 0 - 879 MB | NPU
| SwinV2-Base | TFLITE | float | Qualcomm® QCS8450 | 42.639 ms | 0 - 676 MB | NPU
| SwinV2-Base | TFLITE | float | Qualcomm® Dragonwing™ IQ-9075 | 32.565 ms | 0 - 181 MB | NPU
| SwinV2-Base | TFLITE | float | Qualcomm® Dragonwing™ Q-8750 | 14.878 ms | 0 - 896 MB | NPU
| SwinV2-Base | TFLITE | float | Qualcomm® SA7255P | 71.105 ms | 0 - 885 MB | NPU
| SwinV2-Base | TFLITE | float | Qualcomm® SA8295P | 40.584 ms | 0 - 875 MB | NPU
| SwinV2-Base | TFLITE | float | Snapdragon® 8 Elite Mobile | 14.878 ms | 0 - 896 MB | NPU
| SwinV2-Base | TFLITE | float | Snapdragon® 8 Elite Gen 5 Mobile | 11.386 ms | 0 - 939 MB | NPU

## License
* The license for the original implementation of SwinV2-Base can be found
  [here](https://github.com/pytorch/vision/blob/main/LICENSE).

## References
* [Swin Transformer V2: Scaling Up Capacity and Resolution](https://arxiv.org/abs/2111.09883)
* [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).