--- library_name: pytorch license: other tags: - backbone - bu_auto - real_time - android pipeline_tag: image-classification --- ![](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/mobilenet_v3_small/web-assets/model_demo.png) # MobileNet-v3-Small: Optimized for Qualcomm Devices MobileNetV3Small 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 MobileNet-v3-Small found [here](https://github.com/pytorch/vision/blob/main/torchvision/models/mobilenetv3.py). This repository contains pre-exported model files optimized for Qualcomm® devices. You can use the [Qualcomm® AI Hub Models](https://github.com/quic/ai-hub-models/blob/main/qai_hub_models/models/mobilenet_v3_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 1.24.1 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/mobilenet_v3_small/releases/v0.47.0/mobilenet_v3_small-onnx-float.zip) | ONNX | w8a16 | Universal | QAIRT 2.42, ONNX Runtime 1.24.1 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/mobilenet_v3_small/releases/v0.47.0/mobilenet_v3_small-onnx-w8a16.zip) | QNN_DLC | float | Universal | QAIRT 2.43 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/mobilenet_v3_small/releases/v0.47.0/mobilenet_v3_small-qnn_dlc-float.zip) | QNN_DLC | w8a16 | Universal | QAIRT 2.43 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/mobilenet_v3_small/releases/v0.47.0/mobilenet_v3_small-qnn_dlc-w8a16.zip) | TFLITE | float | Universal | QAIRT 2.43, TFLite 2.17.0 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/mobilenet_v3_small/releases/v0.47.0/mobilenet_v3_small-tflite-float.zip) For more device-specific assets and performance metrics, visit **[MobileNet-v3-Small on Qualcomm® AI Hub](https://aihub.qualcomm.com/models/mobilenet_v3_small)**. ### Option 2: Export with Custom Configurations Use the [Qualcomm® AI Hub Models](https://github.com/quic/ai-hub-models/blob/main/qai_hub_models/models/mobilenet_v3_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 [MobileNet-v3-Small on GitHub](https://github.com/quic/ai-hub-models/blob/main/qai_hub_models/models/mobilenet_v3_small) for usage instructions. ## Model Details **Model Type:** Model_use_case.image_classification **Model Stats:** - Model checkpoint: Imagenet - Input resolution: 224x224 - Number of parameters: 2.54M - Model size (float): 9.71 MB ## Performance Summary | Model | Runtime | Precision | Chipset | Inference Time (ms) | Peak Memory Range (MB) | Primary Compute Unit |---|---|---|---|---|---|--- | MobileNet-v3-Small | ONNX | float | Snapdragon® X Elite | 0.669 ms | 5 - 5 MB | NPU | MobileNet-v3-Small | ONNX | float | Snapdragon® 8 Gen 3 Mobile | 0.355 ms | 0 - 45 MB | NPU | MobileNet-v3-Small | ONNX | float | Qualcomm® QCS8550 (Proxy) | 0.541 ms | 0 - 8 MB | NPU | MobileNet-v3-Small | ONNX | float | Qualcomm® QCS9075 | 0.767 ms | 1 - 3 MB | NPU | MobileNet-v3-Small | ONNX | float | Snapdragon® 8 Elite For Galaxy Mobile | 0.285 ms | 0 - 29 MB | NPU | MobileNet-v3-Small | ONNX | float | Snapdragon® 8 Elite Gen 5 Mobile | 0.244 ms | 0 - 33 MB | NPU | MobileNet-v3-Small | ONNX | float | Snapdragon® X2 Elite | 0.251 ms | 5 - 5 MB | NPU | MobileNet-v3-Small | QNN_DLC | float | Snapdragon® X Elite | 1.012 ms | 1 - 1 MB | NPU | MobileNet-v3-Small | QNN_DLC | float | Snapdragon® 8 Gen 3 Mobile | 0.557 ms | 0 - 46 MB | NPU | MobileNet-v3-Small | QNN_DLC | float | Qualcomm® QCS8275 (Proxy) | 2.114 ms | 1 - 30 MB | NPU | MobileNet-v3-Small | QNN_DLC | float | Qualcomm® QCS8550 (Proxy) | 0.845 ms | 1 - 2 MB | NPU | MobileNet-v3-Small | QNN_DLC | float | Qualcomm® SA8775P | 1.148 ms | 0 - 32 MB | NPU | MobileNet-v3-Small | QNN_DLC | float | Qualcomm® QCS9075 | 0.991 ms | 1 - 3 MB | NPU | MobileNet-v3-Small | QNN_DLC | float | Qualcomm® QCS8450 (Proxy) | 1.587 ms | 0 - 47 MB | NPU | MobileNet-v3-Small | QNN_DLC | float | Qualcomm® SA7255P | 2.114 ms | 1 - 30 MB | NPU | MobileNet-v3-Small | QNN_DLC | float | Qualcomm® SA8295P | 1.497 ms | 0 - 29 MB | NPU | MobileNet-v3-Small | QNN_DLC | float | Snapdragon® 8 Elite For Galaxy Mobile | 0.427 ms | 0 - 30 MB | NPU | MobileNet-v3-Small | QNN_DLC | float | Snapdragon® 8 Elite Gen 5 Mobile | 0.324 ms | 1 - 34 MB | NPU | MobileNet-v3-Small | QNN_DLC | float | Snapdragon® X2 Elite | 0.452 ms | 1 - 1 MB | NPU | MobileNet-v3-Small | QNN_DLC | w8a16 | Snapdragon® X Elite | 0.989 ms | 0 - 0 MB | NPU | MobileNet-v3-Small | QNN_DLC | w8a16 | Snapdragon® 8 Gen 3 Mobile | 0.556 ms | 0 - 37 MB | NPU | MobileNet-v3-Small | QNN_DLC | w8a16 | Qualcomm® QCS6490 | 2.276 ms | 0 - 2 MB | NPU | MobileNet-v3-Small | QNN_DLC | w8a16 | Qualcomm® QCS8275 (Proxy) | 1.7 ms | 0 - 25 MB | NPU | MobileNet-v3-Small | QNN_DLC | w8a16 | Qualcomm® QCS8550 (Proxy) | 0.804 ms | 0 - 2 MB | NPU | MobileNet-v3-Small | QNN_DLC | w8a16 | Qualcomm® SA8775P | 4.299 ms | 0 - 26 MB | NPU | MobileNet-v3-Small | QNN_DLC | w8a16 | Qualcomm® QCS9075 | 0.976 ms | 0 - 2 MB | NPU | MobileNet-v3-Small | QNN_DLC | w8a16 | Qualcomm® QCM6690 | 2.812 ms | 0 - 140 MB | NPU | MobileNet-v3-Small | QNN_DLC | w8a16 | Qualcomm® QCS8450 (Proxy) | 0.986 ms | 0 - 40 MB | NPU | MobileNet-v3-Small | QNN_DLC | w8a16 | Qualcomm® SA7255P | 1.7 ms | 0 - 25 MB | NPU | MobileNet-v3-Small | QNN_DLC | w8a16 | Qualcomm® SA8295P | 1.369 ms | 0 - 23 MB | NPU | MobileNet-v3-Small | QNN_DLC | w8a16 | Snapdragon® 8 Elite For Galaxy Mobile | 0.374 ms | 0 - 29 MB | NPU | MobileNet-v3-Small | QNN_DLC | w8a16 | Snapdragon® 7 Gen 4 Mobile | 0.791 ms | 0 - 25 MB | NPU | MobileNet-v3-Small | QNN_DLC | w8a16 | Snapdragon® 8 Elite Gen 5 Mobile | 0.307 ms | 0 - 28 MB | NPU | MobileNet-v3-Small | QNN_DLC | w8a16 | Snapdragon® X2 Elite | 0.453 ms | 0 - 0 MB | NPU | MobileNet-v3-Small | TFLITE | float | Snapdragon® 8 Gen 3 Mobile | 0.562 ms | 0 - 46 MB | NPU | MobileNet-v3-Small | TFLITE | float | Qualcomm® QCS8275 (Proxy) | 2.201 ms | 0 - 31 MB | NPU | MobileNet-v3-Small | TFLITE | float | Qualcomm® QCS8550 (Proxy) | 0.86 ms | 0 - 2 MB | NPU | MobileNet-v3-Small | TFLITE | float | Qualcomm® SA8775P | 1.184 ms | 0 - 33 MB | NPU | MobileNet-v3-Small | TFLITE | float | Qualcomm® QCS9075 | 1.018 ms | 0 - 8 MB | NPU | MobileNet-v3-Small | TFLITE | float | Qualcomm® QCS8450 (Proxy) | 1.618 ms | 0 - 48 MB | NPU | MobileNet-v3-Small | TFLITE | float | Qualcomm® SA7255P | 2.201 ms | 0 - 31 MB | NPU | MobileNet-v3-Small | TFLITE | float | Qualcomm® SA8295P | 1.504 ms | 0 - 30 MB | NPU | MobileNet-v3-Small | TFLITE | float | Snapdragon® 8 Elite For Galaxy Mobile | 0.436 ms | 0 - 35 MB | NPU | MobileNet-v3-Small | TFLITE | float | Snapdragon® 8 Elite Gen 5 Mobile | 0.344 ms | 0 - 35 MB | NPU ## License * The license for the original implementation of MobileNet-v3-Small can be found [here](https://github.com/pytorch/vision/blob/main/LICENSE). ## References * [Searching for MobileNetV3](https://arxiv.org/abs/1905.02244) * [Source Model Implementation](https://github.com/pytorch/vision/blob/main/torchvision/models/mobilenetv3.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).