| --- |
| library_name: pytorch |
| license: apache-2.0 |
| tags: |
| - backbone |
| - bu_auto |
| - android |
| pipeline_tag: image-classification |
|
|
| --- |
| |
|  |
|
|
| # ConvNextV2-Base: Optimized for Qualcomm Devices |
|
|
| ConvNextV2Base 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 ConvNextV2-Base found [here](https://github.com/huggingface/transformers/blob/main/src/transformers/models/convnextv2/modeling_convnextv2.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/v0.66.0/src/qai_hub_models/models/convnext_v2_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 | |
| |---|---|---|---|---| |
| | ONNX | float | Universal | QAIRT 2.50, ONNX Runtime 1.30.0 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/convnext_v2_base/releases/v0.66.0/convnext_v2_base-onnx-float.zip) |
| | ONNX | w8a16 | Universal | QAIRT 2.50, ONNX Runtime 1.30.0 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/convnext_v2_base/releases/v0.66.0/convnext_v2_base-onnx-w8a16.zip) |
| | QNN_DLC | float | Universal | QAIRT 2.50 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/convnext_v2_base/releases/v0.66.0/convnext_v2_base-qnn_dlc-float.zip) |
| | TFLITE | float | Universal | QAIRT 2.50 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/convnext_v2_base/releases/v0.66.0/convnext_v2_base-tflite-float.zip) |
| |
| For more device-specific assets and performance metrics, visit **[ConvNextV2-Base on Qualcomm® AI Hub](https://aihub.qualcomm.com/models/convnext_v2_base)**. |
| |
| |
| ### Option 2: Export with Custom Configurations |
| |
| Use the [Qualcomm® AI Hub Models](https://github.com/qualcomm/ai-hub-models/blob/v0.66.0/src/qai_hub_models/models/convnext_v2_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 [ConvNextV2-Base on GitHub](https://github.com/qualcomm/ai-hub-models/blob/v0.66.0/src/qai_hub_models/models/convnext_v2_base) for usage instructions. |
| |
| ## Model Details |
| |
| **Model Type:** Model_use_case.image_classification |
|
|
| **Model Stats:** |
| - Input resolution: 224x224 |
| - Model checkpoint: Imagenet |
| - Model size (float): 339 MB |
| - Model size (w8a16): 91.7 MB |
| - Number of parameters: 88.7M |
|
|
| ## Performance Summary |
| | Model | Runtime | Precision | Chipset | Inference Time (ms) | Peak Memory Range (MB) | Primary Compute Unit |
| |---|---|---|---|---|---|--- |
| | ConvNextV2-Base | ONNX | float | Snapdragon® 8 Elite Gen 5 For Galaxy Mobile | 7.536 ms | 0 - 376 MB | NPU |
| | ConvNextV2-Base | ONNX | float | Snapdragon® 8 Elite For Galaxy Mobile | 11.997 ms | 1 - 372 MB | NPU |
| | ConvNextV2-Base | ONNX | float | Snapdragon® X2 Elite | 7.957 ms | 2 - 2 MB | NPU |
| | ConvNextV2-Base | ONNX | float | Snapdragon® X Elite | 17.84 ms | 183 - 183 MB | NPU |
| | ConvNextV2-Base | ONNX | float | Snapdragon® 8 Gen 3 Mobile | 13.038 ms | 0 - 552 MB | NPU |
| | ConvNextV2-Base | ONNX | float | Snapdragon® 8 Gen 1 Mobile | 28.631 ms | 1 - 356 MB | NPU |
| | ConvNextV2-Base | ONNX | float | Qualcomm® Dragonwing™ IQ-8275 | 21.163 ms | 0 - 5 MB | NPU |
| | ConvNextV2-Base | ONNX | float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 17.748 ms | 0 - 194 MB | NPU |
| | ConvNextV2-Base | ONNX | float | Qualcomm® QCS8450 | 28.631 ms | 1 - 356 MB | NPU |
| | ConvNextV2-Base | ONNX | float | Qualcomm® Dragonwing™ IQ-9075 | 20.751 ms | 0 - 4 MB | NPU |
| | ConvNextV2-Base | ONNX | float | Qualcomm® Dragonwing™ IQ-X7181 | 17.84 ms | 183 - 183 MB | NPU |
| | ConvNextV2-Base | ONNX | float | Qualcomm® Dragonwing™ Q-8750 | 11.997 ms | 1 - 372 MB | NPU |
| | ConvNextV2-Base | ONNX | float | Snapdragon® 8 Elite Extreme Gen 6 Mobile | 6.922 ms | 1 - 379 MB | NPU |
| | ConvNextV2-Base | ONNX | float | Snapdragon® 8 Elite Gen 6 Mobile | 8.756 ms | 0 - 408 MB | NPU |
| | ConvNextV2-Base | ONNX | w8a16 | Snapdragon® 8 Elite Gen 5 For Galaxy Mobile | 5.646 ms | 0 - 285 MB | NPU |
| | ConvNextV2-Base | ONNX | w8a16 | Snapdragon® 8 Elite For Galaxy Mobile | 9.614 ms | 0 - 261 MB | NPU |
| | ConvNextV2-Base | ONNX | w8a16 | Snapdragon® X2 Elite | 6.006 ms | 1 - 1 MB | NPU |
| | ConvNextV2-Base | ONNX | w8a16 | Snapdragon® X Elite | 14.205 ms | 92 - 92 MB | NPU |
| | ConvNextV2-Base | ONNX | w8a16 | Snapdragon® 8 Gen 3 Mobile | 9.801 ms | 0 - 336 MB | NPU |
| | ConvNextV2-Base | ONNX | w8a16 | Snapdragon® 8 Gen 1 Mobile | 17.584 ms | 0 - 331 MB | NPU |
| | ConvNextV2-Base | ONNX | w8a16 | Qualcomm® Dragonwing™ IQ-8275 | 13.028 ms | 0 - 5 MB | NPU |
| | ConvNextV2-Base | ONNX | w8a16 | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 13.572 ms | 0 - 99 MB | NPU |
| | ConvNextV2-Base | ONNX | w8a16 | Qualcomm® QCS8450 | 17.584 ms | 0 - 331 MB | NPU |
| | ConvNextV2-Base | ONNX | w8a16 | Qualcomm® Dragonwing™ IQ-9075 | 14.097 ms | 0 - 4 MB | NPU |
| | ConvNextV2-Base | ONNX | w8a16 | Qualcomm® Dragonwing™ IQ-X7181 | 14.205 ms | 92 - 92 MB | NPU |
| | ConvNextV2-Base | ONNX | w8a16 | Qualcomm® Dragonwing™ Q-8750 | 9.614 ms | 0 - 261 MB | NPU |
| | ConvNextV2-Base | ONNX | w8a16 | Snapdragon® 8 Elite Extreme Gen 6 Mobile | 4.627 ms | 0 - 305 MB | NPU |
| | ConvNextV2-Base | ONNX | w8a16 | Snapdragon® 8 Elite Gen 6 Mobile | 6.346 ms | 0 - 289 MB | NPU |
| | ConvNextV2-Base | QNN_DLC | float | Snapdragon® 8 Elite Gen 5 For Galaxy Mobile | 7.64 ms | 0 - 369 MB | NPU |
| | ConvNextV2-Base | QNN_DLC | float | Snapdragon® 8 Elite For Galaxy Mobile | 12.17 ms | 1 - 364 MB | NPU |
| | ConvNextV2-Base | QNN_DLC | float | Snapdragon® X2 Elite | 8.538 ms | 1 - 1 MB | NPU |
| | ConvNextV2-Base | QNN_DLC | float | Snapdragon® X Elite | 18.505 ms | 1 - 1 MB | NPU |
| | ConvNextV2-Base | QNN_DLC | float | Snapdragon® 8 Gen 3 Mobile | 13.285 ms | 1 - 543 MB | NPU |
| | ConvNextV2-Base | QNN_DLC | float | Snapdragon® 8 Gen 1 Mobile | 28.813 ms | 0 - 342 MB | NPU |
| | ConvNextV2-Base | QNN_DLC | float | Qualcomm® Dragonwing™ IQ-8275 | 21.142 ms | 1 - 4 MB | NPU |
| | ConvNextV2-Base | QNN_DLC | float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 17.955 ms | 1 - 3 MB | NPU |
| | ConvNextV2-Base | QNN_DLC | float | Qualcomm® SA8775P | 21.476 ms | 1 - 397 MB | NPU |
| | ConvNextV2-Base | QNN_DLC | float | Qualcomm® SA8650P | 21.476 ms | 1 - 397 MB | NPU |
| | ConvNextV2-Base | QNN_DLC | float | Qualcomm® SA8255P | 21.476 ms | 1 - 397 MB | NPU |
| | ConvNextV2-Base | QNN_DLC | float | Qualcomm® QCS8450 | 28.813 ms | 0 - 342 MB | NPU |
| | ConvNextV2-Base | QNN_DLC | float | Qualcomm® Dragonwing™ IQ-9075 | 20.923 ms | 1 - 3 MB | NPU |
| | ConvNextV2-Base | QNN_DLC | float | Qualcomm® Dragonwing™ IQ-X7181 | 18.505 ms | 1 - 1 MB | NPU |
| | ConvNextV2-Base | QNN_DLC | float | Qualcomm® Dragonwing™ Q-8750 | 12.17 ms | 1 - 364 MB | NPU |
| | ConvNextV2-Base | QNN_DLC | float | Qualcomm® SA7255P | 57.693 ms | 1 - 397 MB | NPU |
| | ConvNextV2-Base | QNN_DLC | float | Qualcomm® SA8295P | 30.498 ms | 1 - 209 MB | NPU |
| | ConvNextV2-Base | QNN_DLC | float | Snapdragon® 8 Elite Extreme Gen 6 Mobile | 6.938 ms | 1 - 370 MB | NPU |
| | ConvNextV2-Base | QNN_DLC | float | Snapdragon® 8 Elite Gen 6 Mobile | 8.895 ms | 1 - 401 MB | NPU |
| | ConvNextV2-Base | TFLITE | float | Snapdragon® 8 Elite Gen 5 For Galaxy Mobile | 5.383 ms | 0 - 324 MB | NPU |
| | ConvNextV2-Base | TFLITE | float | Snapdragon® 8 Elite For Galaxy Mobile | 8.039 ms | 0 - 300 MB | NPU |
| | ConvNextV2-Base | TFLITE | float | Snapdragon® 8 Gen 3 Mobile | 8.229 ms | 0 - 453 MB | NPU |
| | ConvNextV2-Base | TFLITE | float | Snapdragon® 8 Gen 1 Mobile | 26.29 ms | 0 - 361 MB | NPU |
| | ConvNextV2-Base | TFLITE | float | Qualcomm® Dragonwing™ IQ-8275 | 15.814 ms | 0 - 180 MB | NPU |
| | ConvNextV2-Base | TFLITE | float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 11.85 ms | 0 - 3 MB | NPU |
| | ConvNextV2-Base | TFLITE | float | Qualcomm® SA8775P | 15.307 ms | 0 - 301 MB | NPU |
| | ConvNextV2-Base | TFLITE | float | Qualcomm® SA8650P | 15.307 ms | 0 - 301 MB | NPU |
| | ConvNextV2-Base | TFLITE | float | Qualcomm® SA8255P | 15.307 ms | 0 - 301 MB | NPU |
| | ConvNextV2-Base | TFLITE | float | Qualcomm® QCS8450 | 26.29 ms | 0 - 361 MB | NPU |
| | ConvNextV2-Base | TFLITE | float | Qualcomm® Dragonwing™ IQ-9075 | 15.882 ms | 0 - 179 MB | NPU |
| | ConvNextV2-Base | TFLITE | float | Qualcomm® Dragonwing™ Q-8750 | 8.039 ms | 0 - 300 MB | NPU |
| | ConvNextV2-Base | TFLITE | float | Qualcomm® SA7255P | 47.766 ms | 0 - 301 MB | NPU |
| | ConvNextV2-Base | TFLITE | float | Qualcomm® SA8295P | 26.265 ms | 0 - 221 MB | NPU |
| | ConvNextV2-Base | TFLITE | float | Snapdragon® 8 Elite Extreme Gen 6 Mobile | 4.675 ms | 0 - 320 MB | NPU |
| | ConvNextV2-Base | TFLITE | float | Snapdragon® 8 Elite Gen 6 Mobile | 5.191 ms | 0 - 309 MB | NPU |
| |
| ## License |
| * The license for the original implementation of ConvNextV2-Base can be found |
| [here](https://huggingface.co/datasets/choosealicense/licenses/blob/main/markdown/apache-2.0.md). |
| |
| ## References |
| * [ConvNeXt V2: Co-designing and Scaling ConvNets with Masked Autoencoders](https://arxiv.org/abs/2301.00808) |
| * [Source Model Implementation](https://github.com/huggingface/transformers/blob/main/src/transformers/models/convnextv2/modeling_convnextv2.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). |
| |
