--- library_name: pytorch license: other tags: - real_time - android pipeline_tag: object-detection --- ![](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/yolov7/web-assets/model_demo.png) # Yolo-v7: Optimized for Mobile Deployment ## Real-time object detection optimized for mobile and edge YoloV7 is a machine learning model that predicts bounding boxes and classes of objects in an image. This model is an implementation of Yolo-v7 found [here](https://github.com/WongKinYiu/yolov7/). This repository provides scripts to run Yolo-v7 on Qualcomm® devices. More details on model performance across various devices, can be found [here](https://aihub.qualcomm.com/models/yolov7). **WARNING**: The model assets are not readily available for download due to licensing restrictions. ### Model Details - **Model Type:** Model_use_case.object_detection - **Model Stats:** - Model checkpoint: YoloV7 Tiny - Input resolution: 640x640 - Number of parameters: 6.24M - Model size (float): 23.8 MB - Model size (w8a8): 6.23 MB - Model size (w8a16): 6.66 MB | Model | Precision | Device | Chipset | Target Runtime | Inference Time (ms) | Peak Memory Range (MB) | Primary Compute Unit | Target Model |---|---|---|---|---|---|---|---|---| | Yolo-v7 | float | QCS8275 (Proxy) | Qualcomm® QCS8275 (Proxy) | TFLITE | 24.69 ms | 1 - 118 MB | NPU | -- | | Yolo-v7 | float | QCS8275 (Proxy) | Qualcomm® QCS8275 (Proxy) | QNN_DLC | 23.017 ms | 5 - 136 MB | NPU | -- | | Yolo-v7 | float | QCS8450 (Proxy) | Qualcomm® QCS8450 (Proxy) | TFLITE | 12.899 ms | 1 - 48 MB | NPU | -- | | Yolo-v7 | float | QCS8450 (Proxy) | Qualcomm® QCS8450 (Proxy) | QNN_DLC | 13.272 ms | 5 - 43 MB | NPU | -- | | Yolo-v7 | float | QCS8550 (Proxy) | Qualcomm® QCS8550 (Proxy) | TFLITE | 9.495 ms | 0 - 100 MB | NPU | -- | | Yolo-v7 | float | QCS8550 (Proxy) | Qualcomm® QCS8550 (Proxy) | QNN_DLC | 10.461 ms | 3 - 32 MB | NPU | -- | | Yolo-v7 | float | QCS8550 (Proxy) | Qualcomm® QCS8550 (Proxy) | ONNX | 10.52 ms | 0 - 48 MB | NPU | -- | | Yolo-v7 | float | QCS9075 (Proxy) | Qualcomm® QCS9075 (Proxy) | TFLITE | 11.354 ms | 1 - 117 MB | NPU | -- | | Yolo-v7 | float | QCS9075 (Proxy) | Qualcomm® QCS9075 (Proxy) | QNN_DLC | 10.502 ms | 2 - 130 MB | NPU | -- | | Yolo-v7 | float | SA7255P ADP | Qualcomm® SA7255P | TFLITE | 24.69 ms | 1 - 118 MB | NPU | -- | | Yolo-v7 | float | SA7255P ADP | Qualcomm® SA7255P | QNN_DLC | 23.017 ms | 5 - 136 MB | NPU | -- | | Yolo-v7 | float | SA8255 (Proxy) | Qualcomm® SA8255P (Proxy) | TFLITE | 9.471 ms | 0 - 101 MB | NPU | -- | | Yolo-v7 | float | SA8255 (Proxy) | Qualcomm® SA8255P (Proxy) | QNN_DLC | 10.432 ms | 5 - 21 MB | NPU | -- | | Yolo-v7 | float | SA8295P ADP | Qualcomm® SA8295P | TFLITE | 14.446 ms | 1 - 40 MB | NPU | -- | | Yolo-v7 | float | SA8295P ADP | Qualcomm® SA8295P | QNN_DLC | 11.804 ms | 2 - 42 MB | NPU | -- | | Yolo-v7 | float | SA8650 (Proxy) | Qualcomm® SA8650P (Proxy) | TFLITE | 9.481 ms | 0 - 102 MB | NPU | -- | | Yolo-v7 | float | SA8650 (Proxy) | Qualcomm® SA8650P (Proxy) | QNN_DLC | 10.461 ms | 5 - 21 MB | NPU | -- | | Yolo-v7 | float | SA8775P ADP | Qualcomm® SA8775P | TFLITE | 11.354 ms | 1 - 117 MB | NPU | -- | | Yolo-v7 | float | SA8775P ADP | Qualcomm® SA8775P | QNN_DLC | 10.502 ms | 2 - 130 MB | NPU | -- | | Yolo-v7 | float | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 Mobile | TFLITE | 9.408 ms | 0 - 103 MB | NPU | -- | | Yolo-v7 | float | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 Mobile | QNN_DLC | 10.539 ms | 5 - 21 MB | NPU | -- | | Yolo-v7 | float | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 Mobile | ONNX | 10.612 ms | 0 - 40 MB | NPU | -- | | Yolo-v7 | float | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | TFLITE | 6.7 ms | 8 - 221 MB | NPU | -- | | Yolo-v7 | float | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | QNN_DLC | 6.008 ms | 5 - 313 MB | NPU | -- | | Yolo-v7 | float | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | ONNX | 7.683 ms | 5 - 227 MB | NPU | -- | | Yolo-v7 | float | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite Mobile | TFLITE | 7.059 ms | 1 - 116 MB | NPU | -- | | Yolo-v7 | float | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite Mobile | QNN_DLC | 5.878 ms | 5 - 128 MB | NPU | -- | | Yolo-v7 | float | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite Mobile | ONNX | 6.636 ms | 5 - 136 MB | NPU | -- | | Yolo-v7 | float | Snapdragon X Elite CRD | Snapdragon® X Elite | QNN_DLC | 11.142 ms | 210 - 210 MB | NPU | -- | | Yolo-v7 | float | Snapdragon X Elite CRD | Snapdragon® X Elite | ONNX | 12.352 ms | 8 - 8 MB | NPU | -- | | Yolo-v7 | w8a16 | QCS8275 (Proxy) | Qualcomm® QCS8275 (Proxy) | QNN_DLC | 18.123 ms | 2 - 37 MB | NPU | -- | | Yolo-v7 | w8a16 | QCS8450 (Proxy) | Qualcomm® QCS8450 (Proxy) | QNN_DLC | 12.229 ms | 2 - 54 MB | NPU | -- | | Yolo-v7 | w8a16 | QCS8550 (Proxy) | Qualcomm® QCS8550 (Proxy) | QNN_DLC | 9.433 ms | 2 - 16 MB | NPU | -- | | Yolo-v7 | w8a16 | QCS8550 (Proxy) | Qualcomm® QCS8550 (Proxy) | ONNX | 120.182 ms | 17 - 382 MB | NPU | -- | | Yolo-v7 | w8a16 | QCS9075 (Proxy) | Qualcomm® QCS9075 (Proxy) | QNN_DLC | 10.283 ms | 2 - 39 MB | NPU | -- | | Yolo-v7 | w8a16 | RB3 Gen 2 (Proxy) | Qualcomm® QCS6490 (Proxy) | QNN_DLC | 23.85 ms | 2 - 43 MB | NPU | -- | | Yolo-v7 | w8a16 | RB3 Gen 2 (Proxy) | Qualcomm® QCS6490 (Proxy) | ONNX | 210.86 ms | 85 - 100 MB | CPU | -- | | Yolo-v7 | w8a16 | RB5 (Proxy) | Qualcomm® QCS8250 (Proxy) | ONNX | 184.504 ms | 77 - 88 MB | CPU | -- | | Yolo-v7 | w8a16 | SA7255P ADP | Qualcomm® SA7255P | QNN_DLC | 18.123 ms | 2 - 37 MB | NPU | -- | | Yolo-v7 | w8a16 | SA8255 (Proxy) | Qualcomm® SA8255P (Proxy) | QNN_DLC | 9.405 ms | 2 - 15 MB | NPU | -- | | Yolo-v7 | w8a16 | SA8295P ADP | Qualcomm® SA8295P | QNN_DLC | 11.982 ms | 2 - 50 MB | NPU | -- | | Yolo-v7 | w8a16 | SA8650 (Proxy) | Qualcomm® SA8650P (Proxy) | QNN_DLC | 9.433 ms | 2 - 14 MB | NPU | -- | | Yolo-v7 | w8a16 | SA8775P ADP | Qualcomm® SA8775P | QNN_DLC | 10.283 ms | 2 - 39 MB | NPU | -- | | Yolo-v7 | w8a16 | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 Mobile | QNN_DLC | 9.443 ms | 2 - 18 MB | NPU | -- | | Yolo-v7 | w8a16 | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 Mobile | ONNX | 122.948 ms | 43 - 373 MB | NPU | -- | | Yolo-v7 | w8a16 | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | QNN_DLC | 5.124 ms | 2 - 53 MB | NPU | -- | | Yolo-v7 | w8a16 | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | ONNX | 92.731 ms | 37 - 1273 MB | NPU | -- | | Yolo-v7 | w8a16 | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite Mobile | QNN_DLC | 4.979 ms | 2 - 46 MB | NPU | -- | | Yolo-v7 | w8a16 | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite Mobile | ONNX | 91.176 ms | 54 - 1814 MB | NPU | -- | | Yolo-v7 | w8a16 | Snapdragon X Elite CRD | Snapdragon® X Elite | QNN_DLC | 7.915 ms | 12 - 12 MB | NPU | -- | | Yolo-v7 | w8a16 | Snapdragon X Elite CRD | Snapdragon® X Elite | ONNX | 116.555 ms | 58 - 58 MB | NPU | -- | | Yolo-v7 | w8a8 | QCS8275 (Proxy) | Qualcomm® QCS8275 (Proxy) | TFLITE | 5.567 ms | 0 - 27 MB | NPU | -- | | Yolo-v7 | w8a8 | QCS8275 (Proxy) | Qualcomm® QCS8275 (Proxy) | QNN_DLC | 4.303 ms | 1 - 30 MB | NPU | -- | | Yolo-v7 | w8a8 | QCS8450 (Proxy) | Qualcomm® QCS8450 (Proxy) | TFLITE | 3.315 ms | 0 - 49 MB | NPU | -- | | Yolo-v7 | w8a8 | QCS8450 (Proxy) | Qualcomm® QCS8450 (Proxy) | QNN_DLC | 2.742 ms | 1 - 47 MB | NPU | -- | | Yolo-v7 | w8a8 | QCS8550 (Proxy) | Qualcomm® QCS8550 (Proxy) | TFLITE | 2.617 ms | 0 - 32 MB | NPU | -- | | Yolo-v7 | w8a8 | QCS8550 (Proxy) | Qualcomm® QCS8550 (Proxy) | QNN_DLC | 1.914 ms | 1 - 11 MB | NPU | -- | | Yolo-v7 | w8a8 | QCS8550 (Proxy) | Qualcomm® QCS8550 (Proxy) | ONNX | 3.839 ms | 0 - 93 MB | NPU | -- | | Yolo-v7 | w8a8 | QCS9075 (Proxy) | Qualcomm® QCS9075 (Proxy) | TFLITE | 3.053 ms | 0 - 27 MB | NPU | -- | | Yolo-v7 | w8a8 | QCS9075 (Proxy) | Qualcomm® QCS9075 (Proxy) | QNN_DLC | 2.361 ms | 1 - 31 MB | NPU | -- | | Yolo-v7 | w8a8 | RB3 Gen 2 (Proxy) | Qualcomm® QCS6490 (Proxy) | TFLITE | 19.88 ms | 8 - 56 MB | NPU | -- | | Yolo-v7 | w8a8 | RB3 Gen 2 (Proxy) | Qualcomm® QCS6490 (Proxy) | QNN_DLC | 8.683 ms | 1 - 38 MB | NPU | -- | | Yolo-v7 | w8a8 | RB3 Gen 2 (Proxy) | Qualcomm® QCS6490 (Proxy) | ONNX | 57.963 ms | 38 - 54 MB | CPU | -- | | Yolo-v7 | w8a8 | RB5 (Proxy) | Qualcomm® QCS8250 (Proxy) | TFLITE | 114.496 ms | 18 - 48 MB | GPU | -- | | Yolo-v7 | w8a8 | RB5 (Proxy) | Qualcomm® QCS8250 (Proxy) | ONNX | 50.506 ms | 35 - 46 MB | CPU | -- | | Yolo-v7 | w8a8 | SA7255P ADP | Qualcomm® SA7255P | TFLITE | 5.567 ms | 0 - 27 MB | NPU | -- | | Yolo-v7 | w8a8 | SA7255P ADP | Qualcomm® SA7255P | QNN_DLC | 4.303 ms | 1 - 30 MB | NPU | -- | | Yolo-v7 | w8a8 | SA8255 (Proxy) | Qualcomm® SA8255P (Proxy) | TFLITE | 2.621 ms | 0 - 8 MB | NPU | -- | | Yolo-v7 | w8a8 | SA8255 (Proxy) | Qualcomm® SA8255P (Proxy) | QNN_DLC | 1.918 ms | 1 - 14 MB | NPU | -- | | Yolo-v7 | w8a8 | SA8295P ADP | Qualcomm® SA8295P | TFLITE | 4.344 ms | 0 - 35 MB | NPU | -- | | Yolo-v7 | w8a8 | SA8295P ADP | Qualcomm® SA8295P | QNN_DLC | 3.251 ms | 1 - 37 MB | NPU | -- | | Yolo-v7 | w8a8 | SA8650 (Proxy) | Qualcomm® SA8650P (Proxy) | TFLITE | 2.626 ms | 0 - 33 MB | NPU | -- | | Yolo-v7 | w8a8 | SA8650 (Proxy) | Qualcomm® SA8650P (Proxy) | QNN_DLC | 1.912 ms | 0 - 9 MB | NPU | -- | | Yolo-v7 | w8a8 | SA8775P ADP | Qualcomm® SA8775P | TFLITE | 3.053 ms | 0 - 27 MB | NPU | -- | | Yolo-v7 | w8a8 | SA8775P ADP | Qualcomm® SA8775P | QNN_DLC | 2.361 ms | 1 - 31 MB | NPU | -- | | Yolo-v7 | w8a8 | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 Mobile | TFLITE | 2.668 ms | 0 - 22 MB | NPU | -- | | Yolo-v7 | w8a8 | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 Mobile | QNN_DLC | 1.915 ms | 1 - 13 MB | NPU | -- | | Yolo-v7 | w8a8 | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 Mobile | ONNX | 3.864 ms | 0 - 103 MB | NPU | -- | | Yolo-v7 | w8a8 | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | TFLITE | 1.752 ms | 0 - 40 MB | NPU | -- | | Yolo-v7 | w8a8 | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | QNN_DLC | 1.287 ms | 1 - 47 MB | NPU | -- | | Yolo-v7 | w8a8 | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | ONNX | 2.718 ms | 0 - 348 MB | NPU | -- | | Yolo-v7 | w8a8 | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite Mobile | TFLITE | 1.608 ms | 0 - 29 MB | NPU | -- | | Yolo-v7 | w8a8 | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite Mobile | QNN_DLC | 1.175 ms | 1 - 37 MB | NPU | -- | | Yolo-v7 | w8a8 | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite Mobile | ONNX | 2.567 ms | 1 - 131 MB | NPU | -- | | Yolo-v7 | w8a8 | Snapdragon X Elite CRD | Snapdragon® X Elite | QNN_DLC | 2.188 ms | 22 - 22 MB | NPU | -- | | Yolo-v7 | w8a8 | Snapdragon X Elite CRD | Snapdragon® X Elite | ONNX | 3.848 ms | 6 - 6 MB | NPU | -- | ## Installation Install the package via pip: ```bash pip install "qai-hub-models[yolov7]" ``` ## Configure Qualcomm® AI Hub to run this model on a cloud-hosted device Sign-in to [Qualcomm® AI Hub](https://app.aihub.qualcomm.com/) with your Qualcomm® ID. Once signed in navigate to `Account -> Settings -> API Token`. With this API token, you can configure your client to run models on the cloud hosted devices. ```bash qai-hub configure --api_token API_TOKEN ``` Navigate to [docs](https://app.aihub.qualcomm.com/docs/) for more information. ## Demo off target The package contains a simple end-to-end demo that downloads pre-trained weights and runs this model on a sample input. ```bash python -m qai_hub_models.models.yolov7.demo ``` The above demo runs a reference implementation of pre-processing, model inference, and post processing. **NOTE**: If you want running in a Jupyter Notebook or Google Colab like environment, please add the following to your cell (instead of the above). ``` %run -m qai_hub_models.models.yolov7.demo ``` ### Run model on a cloud-hosted device In addition to the demo, you can also run the model on a cloud-hosted Qualcomm® device. This script does the following: * Performance check on-device on a cloud-hosted device * Downloads compiled assets that can be deployed on-device for Android. * Accuracy check between PyTorch and on-device outputs. ```bash python -m qai_hub_models.models.yolov7.export ``` ## How does this work? This [export script](https://aihub.qualcomm.com/models/yolov7/qai_hub_models/models/Yolo-v7/export.py) leverages [Qualcomm® AI Hub](https://aihub.qualcomm.com/) to optimize, validate, and deploy this model on-device. Lets go through each step below in detail: Step 1: **Compile model for on-device deployment** To compile a PyTorch model for on-device deployment, we first trace the model in memory using the `jit.trace` and then call the `submit_compile_job` API. ```python import torch import qai_hub as hub from qai_hub_models.models.yolov7 import Model # Load the model torch_model = Model.from_pretrained() # Device device = hub.Device("Samsung Galaxy S24") # Trace model input_shape = torch_model.get_input_spec() sample_inputs = torch_model.sample_inputs() pt_model = torch.jit.trace(torch_model, [torch.tensor(data[0]) for _, data in sample_inputs.items()]) # Compile model on a specific device compile_job = hub.submit_compile_job( model=pt_model, device=device, input_specs=torch_model.get_input_spec(), ) # Get target model to run on-device target_model = compile_job.get_target_model() ``` Step 2: **Performance profiling on cloud-hosted device** After compiling models from step 1. Models can be profiled model on-device using the `target_model`. Note that this scripts runs the model on a device automatically provisioned in the cloud. Once the job is submitted, you can navigate to a provided job URL to view a variety of on-device performance metrics. ```python profile_job = hub.submit_profile_job( model=target_model, device=device, ) ``` Step 3: **Verify on-device accuracy** To verify the accuracy of the model on-device, you can run on-device inference on sample input data on the same cloud hosted device. ```python input_data = torch_model.sample_inputs() inference_job = hub.submit_inference_job( model=target_model, device=device, inputs=input_data, ) on_device_output = inference_job.download_output_data() ``` With the output of the model, you can compute like PSNR, relative errors or spot check the output with expected output. **Note**: This on-device profiling and inference requires access to Qualcomm® AI Hub. [Sign up for access](https://myaccount.qualcomm.com/signup). ## Run demo on a cloud-hosted device You can also run the demo on-device. ```bash python -m qai_hub_models.models.yolov7.demo --eval-mode on-device ``` **NOTE**: If you want running in a Jupyter Notebook or Google Colab like environment, please add the following to your cell (instead of the above). ``` %run -m qai_hub_models.models.yolov7.demo -- --eval-mode on-device ``` ## Deploying compiled model to Android The models can be deployed using multiple runtimes: - TensorFlow Lite (`.tflite` export): [This tutorial](https://www.tensorflow.org/lite/android/quickstart) provides a guide to deploy the .tflite model in an Android application. - QNN (`.so` export ): This [sample app](https://docs.qualcomm.com/bundle/publicresource/topics/80-63442-50/sample_app.html) provides instructions on how to use the `.so` shared library in an Android application. ## View on Qualcomm® AI Hub Get more details on Yolo-v7's performance across various devices [here](https://aihub.qualcomm.com/models/yolov7). Explore all available models on [Qualcomm® AI Hub](https://aihub.qualcomm.com/) ## License * The license for the original implementation of Yolo-v7 can be found [here](https://github.com/WongKinYiu/yolov7/blob/main/LICENSE.md). * The license for the compiled assets for on-device deployment can be found [here](https://github.com/WongKinYiu/yolov7/blob/main/LICENSE.md) ## References * [YOLOv7: Trainable bag-of-freebies sets new state-of-the-art for real-time object detectors](https://arxiv.org/abs/2207.02696) * [Source Model Implementation](https://github.com/WongKinYiu/yolov7/) ## 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).