--- library_name: pytorch license: other tags: - android pipeline_tag: image-classification --- ![](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/regnet_y_800mf/web-assets/model_demo.png) # RegNet-Y-800MF: Optimized for Qualcomm Devices RegNet_Y_800MF is part of the RegNet family of models designed for efficient and scalable image classification. It uses a simple yet effective design space to balance performance and computational cost, making it suitable for mobile and edge devices. 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/regnet_y_800mf) 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/regnet_y_800mf/releases/v0.51.0/regnet_y_800mf-onnx-float.zip) | ONNX | w8a8 | Universal | QAIRT 2.42, ONNX Runtime 1.24.3 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/regnet_y_800mf/releases/v0.51.0/regnet_y_800mf-onnx-w8a8.zip) | QNN_DLC | float | Universal | QAIRT 2.45 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/regnet_y_800mf/releases/v0.51.0/regnet_y_800mf-qnn_dlc-float.zip) | QNN_DLC | w8a8 | Universal | QAIRT 2.45 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/regnet_y_800mf/releases/v0.51.0/regnet_y_800mf-qnn_dlc-w8a8.zip) | TFLITE | float | Universal | QAIRT 2.45 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/regnet_y_800mf/releases/v0.51.0/regnet_y_800mf-tflite-float.zip) | TFLITE | w8a8 | Universal | QAIRT 2.45 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/regnet_y_800mf/releases/v0.51.0/regnet_y_800mf-tflite-w8a8.zip) For more device-specific assets and performance metrics, visit **[RegNet-Y-800MF on Qualcomm® AI Hub](https://aihub.qualcomm.com/models/regnet_y_800mf)**. ### 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/regnet_y_800mf) 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 [RegNet-Y-800MF on GitHub](https://github.com/qualcomm/ai-hub-models/blob/main/src/qai_hub_models/models/regnet_y_800mf) for usage instructions. ## Model Details **Model Type:** Model_use_case.image_classification **Model Stats:** - Model checkpoint: regnet_y_800mf-1b27b58c.pth - Input resolution: 1x3x224 - Model size: ~6.3 MB ## Performance Summary | Model | Runtime | Precision | Chipset | Inference Time (ms) | Peak Memory Range (MB) | Primary Compute Unit |---|---|---|---|---|---|--- | RegNet-Y-800MF | ONNX | float | Snapdragon® 8 Elite Gen 5 Mobile | 0.534 ms | 0 - 61 MB | NPU | RegNet-Y-800MF | ONNX | float | Snapdragon® X2 Elite | 0.55 ms | 15 - 15 MB | NPU | RegNet-Y-800MF | ONNX | float | Snapdragon® X Elite | 1.306 ms | 14 - 14 MB | NPU | RegNet-Y-800MF | ONNX | float | Snapdragon® 8 Gen 3 Mobile | 0.71 ms | 0 - 77 MB | NPU | RegNet-Y-800MF | ONNX | float | Qualcomm® QCS8550 (Proxy) | 1.089 ms | 0 - 18 MB | NPU | RegNet-Y-800MF | ONNX | float | Qualcomm® QCS9075 | 1.484 ms | 1 - 3 MB | NPU | RegNet-Y-800MF | ONNX | float | Snapdragon® 8 Elite For Galaxy Mobile | 0.557 ms | 0 - 63 MB | NPU | RegNet-Y-800MF | ONNX | w8a8 | Snapdragon® 8 Elite Gen 5 Mobile | 0.387 ms | 0 - 59 MB | NPU | RegNet-Y-800MF | ONNX | w8a8 | Snapdragon® X2 Elite | 0.384 ms | 7 - 7 MB | NPU | RegNet-Y-800MF | ONNX | w8a8 | Snapdragon® X Elite | 0.876 ms | 7 - 7 MB | NPU | RegNet-Y-800MF | ONNX | w8a8 | Snapdragon® 8 Gen 3 Mobile | 0.507 ms | 0 - 86 MB | NPU | RegNet-Y-800MF | ONNX | w8a8 | Qualcomm® QCS6490 | 15.29 ms | 5 - 11 MB | CPU | RegNet-Y-800MF | ONNX | w8a8 | Qualcomm® QCS8550 (Proxy) | 0.695 ms | 0 - 12 MB | NPU | RegNet-Y-800MF | ONNX | w8a8 | Qualcomm® QCS9075 | 0.89 ms | 0 - 3 MB | NPU | RegNet-Y-800MF | ONNX | w8a8 | Qualcomm® QCM6690 | 8.667 ms | 6 - 15 MB | CPU | RegNet-Y-800MF | ONNX | w8a8 | Snapdragon® 8 Elite For Galaxy Mobile | 0.436 ms | 0 - 48 MB | NPU | RegNet-Y-800MF | ONNX | w8a8 | Snapdragon® 7 Gen 4 Mobile | 6.392 ms | 5 - 14 MB | CPU | RegNet-Y-800MF | QNN_DLC | float | Snapdragon® 8 Elite Gen 5 Mobile | 0.61 ms | 1 - 55 MB | NPU | RegNet-Y-800MF | QNN_DLC | float | Snapdragon® X2 Elite | 0.816 ms | 1 - 1 MB | NPU | RegNet-Y-800MF | QNN_DLC | float | Snapdragon® X Elite | 1.626 ms | 1 - 1 MB | NPU | RegNet-Y-800MF | QNN_DLC | float | Snapdragon® 8 Gen 3 Mobile | 0.913 ms | 0 - 74 MB | NPU | RegNet-Y-800MF | QNN_DLC | float | Qualcomm® QCS8275 (Proxy) | 4.053 ms | 1 - 50 MB | NPU | RegNet-Y-800MF | QNN_DLC | float | Qualcomm® QCS8550 (Proxy) | 1.411 ms | 0 - 14 MB | NPU | RegNet-Y-800MF | QNN_DLC | float | Qualcomm® SA8775P | 1.901 ms | 1 - 53 MB | NPU | RegNet-Y-800MF | QNN_DLC | float | Qualcomm® QCS9075 | 1.702 ms | 1 - 3 MB | NPU | RegNet-Y-800MF | QNN_DLC | float | Qualcomm® QCS8450 (Proxy) | 2.245 ms | 0 - 69 MB | NPU | RegNet-Y-800MF | QNN_DLC | float | Qualcomm® SA7255P | 4.053 ms | 1 - 50 MB | NPU | RegNet-Y-800MF | QNN_DLC | float | Qualcomm® SA8295P | 2.135 ms | 0 - 45 MB | NPU | RegNet-Y-800MF | QNN_DLC | float | Snapdragon® 8 Elite For Galaxy Mobile | 0.692 ms | 0 - 51 MB | NPU | RegNet-Y-800MF | QNN_DLC | w8a8 | Snapdragon® 8 Elite Gen 5 Mobile | 0.308 ms | 0 - 50 MB | NPU | RegNet-Y-800MF | QNN_DLC | w8a8 | Snapdragon® X2 Elite | 0.443 ms | 0 - 0 MB | NPU | RegNet-Y-800MF | QNN_DLC | w8a8 | Snapdragon® X Elite | 0.86 ms | 0 - 0 MB | NPU | RegNet-Y-800MF | QNN_DLC | w8a8 | Snapdragon® 8 Gen 3 Mobile | 0.484 ms | 0 - 65 MB | NPU | RegNet-Y-800MF | QNN_DLC | w8a8 | Qualcomm® QCS6490 | 1.683 ms | 2 - 4 MB | NPU | RegNet-Y-800MF | QNN_DLC | w8a8 | Qualcomm® QCS8275 (Proxy) | 1.563 ms | 0 - 47 MB | NPU | RegNet-Y-800MF | QNN_DLC | w8a8 | Qualcomm® QCS8550 (Proxy) | 0.701 ms | 0 - 68 MB | NPU | RegNet-Y-800MF | QNN_DLC | w8a8 | Qualcomm® SA8775P | 0.92 ms | 0 - 50 MB | NPU | RegNet-Y-800MF | QNN_DLC | w8a8 | Qualcomm® QCS9075 | 0.84 ms | 0 - 2 MB | NPU | RegNet-Y-800MF | QNN_DLC | w8a8 | Qualcomm® QCM6690 | 2.815 ms | 0 - 48 MB | NPU | RegNet-Y-800MF | QNN_DLC | w8a8 | Qualcomm® QCS8450 (Proxy) | 0.862 ms | 0 - 66 MB | NPU | RegNet-Y-800MF | QNN_DLC | w8a8 | Qualcomm® SA7255P | 1.563 ms | 0 - 47 MB | NPU | RegNet-Y-800MF | QNN_DLC | w8a8 | Qualcomm® SA8295P | 1.106 ms | 0 - 45 MB | NPU | RegNet-Y-800MF | QNN_DLC | w8a8 | Snapdragon® 8 Elite For Galaxy Mobile | 0.373 ms | 0 - 50 MB | NPU | RegNet-Y-800MF | QNN_DLC | w8a8 | Snapdragon® 7 Gen 4 Mobile | 0.722 ms | 0 - 46 MB | NPU | RegNet-Y-800MF | TFLITE | float | Snapdragon® 8 Elite Gen 5 Mobile | 0.608 ms | 0 - 62 MB | NPU | RegNet-Y-800MF | TFLITE | float | Snapdragon® 8 Gen 3 Mobile | 0.911 ms | 0 - 86 MB | NPU | RegNet-Y-800MF | TFLITE | float | Qualcomm® QCS8275 (Proxy) | 4.058 ms | 0 - 58 MB | NPU | RegNet-Y-800MF | TFLITE | float | Qualcomm® QCS8550 (Proxy) | 1.399 ms | 0 - 2 MB | NPU | RegNet-Y-800MF | TFLITE | float | Qualcomm® SA8775P | 1.904 ms | 0 - 63 MB | NPU | RegNet-Y-800MF | TFLITE | float | Qualcomm® QCS9075 | 1.7 ms | 0 - 17 MB | NPU | RegNet-Y-800MF | TFLITE | float | Qualcomm® QCS8450 (Proxy) | 2.209 ms | 0 - 74 MB | NPU | RegNet-Y-800MF | TFLITE | float | Qualcomm® SA7255P | 4.058 ms | 0 - 58 MB | NPU | RegNet-Y-800MF | TFLITE | float | Qualcomm® SA8295P | 2.139 ms | 0 - 50 MB | NPU | RegNet-Y-800MF | TFLITE | float | Snapdragon® 8 Elite For Galaxy Mobile | 0.689 ms | 0 - 62 MB | NPU | RegNet-Y-800MF | TFLITE | w8a8 | Snapdragon® 8 Elite Gen 5 Mobile | 0.252 ms | 0 - 51 MB | NPU | RegNet-Y-800MF | TFLITE | w8a8 | Snapdragon® 8 Gen 3 Mobile | 0.356 ms | 0 - 66 MB | NPU | RegNet-Y-800MF | TFLITE | w8a8 | Qualcomm® QCS6490 | 1.265 ms | 0 - 9 MB | NPU | RegNet-Y-800MF | TFLITE | w8a8 | Qualcomm® QCS8275 (Proxy) | 1.268 ms | 0 - 47 MB | NPU | RegNet-Y-800MF | TFLITE | w8a8 | Qualcomm® QCS8550 (Proxy) | 0.514 ms | 0 - 2 MB | NPU | RegNet-Y-800MF | TFLITE | w8a8 | Qualcomm® SA8775P | 0.741 ms | 0 - 51 MB | NPU | RegNet-Y-800MF | TFLITE | w8a8 | Qualcomm® QCS9075 | 0.664 ms | 0 - 10 MB | NPU | RegNet-Y-800MF | TFLITE | w8a8 | Qualcomm® QCM6690 | 2.302 ms | 0 - 44 MB | NPU | RegNet-Y-800MF | TFLITE | w8a8 | Qualcomm® QCS8450 (Proxy) | 0.679 ms | 0 - 66 MB | NPU | RegNet-Y-800MF | TFLITE | w8a8 | Qualcomm® SA7255P | 1.268 ms | 0 - 47 MB | NPU | RegNet-Y-800MF | TFLITE | w8a8 | Qualcomm® SA8295P | 0.913 ms | 0 - 44 MB | NPU | RegNet-Y-800MF | TFLITE | w8a8 | Snapdragon® 8 Elite For Galaxy Mobile | 0.302 ms | 0 - 43 MB | NPU | RegNet-Y-800MF | TFLITE | w8a8 | Snapdragon® 7 Gen 4 Mobile | 0.54 ms | 0 - 43 MB | NPU ## License * The license for the original implementation of RegNet-Y-800MF can be found [here](https://github.com/pytorch/vision/blob/main/LICENSE). ## 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).