--- 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/yolox/web-assets/model_demo.png) # Yolo-X: Optimized for Mobile Deployment ## Real-time object detection optimized for mobile and edge YoloX is a machine learning model that predicts bounding boxes and classes of objects in an image. This model is an implementation of Yolo-X found [here](https://github.com/Megvii-BaseDetection/YOLOX/). This repository provides scripts to run Yolo-X on Qualcomm® devices. More details on model performance across various devices, can be found [here](https://aihub.qualcomm.com/models/yolox). ### Model Details - **Model Type:** Model_use_case.object_detection - **Model Stats:** - Model checkpoint: YoloX Medium - Input resolution: 640x640 - Number of parameters: 25.3M - Model size (float): 96.7 MB | Model | Precision | Device | Chipset | Target Runtime | Inference Time (ms) | Peak Memory Range (MB) | Primary Compute Unit | Target Model |---|---|---|---|---|---|---|---|---| | Yolo-X | float | QCS8275 (Proxy) | Qualcomm® QCS8275 (Proxy) | TFLITE | 79.809 ms | 0 - 75 MB | NPU | [Yolo-X.tflite](https://huggingface.co/qualcomm/Yolo-X/blob/main/Yolo-X.tflite) | | Yolo-X | float | QCS8275 (Proxy) | Qualcomm® QCS8275 (Proxy) | QNN | 75.122 ms | 1 - 11 MB | NPU | Use Export Script | | Yolo-X | float | QCS8450 (Proxy) | Qualcomm® QCS8450 (Proxy) | TFLITE | 34.789 ms | 0 - 106 MB | NPU | [Yolo-X.tflite](https://huggingface.co/qualcomm/Yolo-X/blob/main/Yolo-X.tflite) | | Yolo-X | float | QCS8450 (Proxy) | Qualcomm® QCS8450 (Proxy) | QNN | 34.374 ms | 5 - 52 MB | NPU | Use Export Script | | Yolo-X | float | QCS8550 (Proxy) | Qualcomm® QCS8550 (Proxy) | TFLITE | 20.102 ms | 0 - 18 MB | NPU | [Yolo-X.tflite](https://huggingface.co/qualcomm/Yolo-X/blob/main/Yolo-X.tflite) | | Yolo-X | float | QCS8550 (Proxy) | Qualcomm® QCS8550 (Proxy) | QNN | 15.609 ms | 5 - 8 MB | NPU | Use Export Script | | Yolo-X | float | QCS9075 (Proxy) | Qualcomm® QCS9075 (Proxy) | TFLITE | 26.709 ms | 0 - 76 MB | NPU | [Yolo-X.tflite](https://huggingface.co/qualcomm/Yolo-X/blob/main/Yolo-X.tflite) | | Yolo-X | float | QCS9075 (Proxy) | Qualcomm® QCS9075 (Proxy) | QNN | 22.043 ms | 0 - 13 MB | NPU | Use Export Script | | Yolo-X | float | SA7255P ADP | Qualcomm® SA7255P | TFLITE | 79.809 ms | 0 - 75 MB | NPU | [Yolo-X.tflite](https://huggingface.co/qualcomm/Yolo-X/blob/main/Yolo-X.tflite) | | Yolo-X | float | SA7255P ADP | Qualcomm® SA7255P | QNN | 75.122 ms | 1 - 11 MB | NPU | Use Export Script | | Yolo-X | float | SA8255 (Proxy) | Qualcomm® SA8255P (Proxy) | TFLITE | 20.353 ms | 0 - 16 MB | NPU | [Yolo-X.tflite](https://huggingface.co/qualcomm/Yolo-X/blob/main/Yolo-X.tflite) | | Yolo-X | float | SA8255 (Proxy) | Qualcomm® SA8255P (Proxy) | QNN | 15.579 ms | 5 - 7 MB | NPU | Use Export Script | | Yolo-X | float | SA8295P ADP | Qualcomm® SA8295P | TFLITE | 37.164 ms | 0 - 68 MB | NPU | [Yolo-X.tflite](https://huggingface.co/qualcomm/Yolo-X/blob/main/Yolo-X.tflite) | | Yolo-X | float | SA8295P ADP | Qualcomm® SA8295P | QNN | 28.667 ms | 0 - 18 MB | NPU | Use Export Script | | Yolo-X | float | SA8650 (Proxy) | Qualcomm® SA8650P (Proxy) | TFLITE | 20.652 ms | 0 - 17 MB | NPU | [Yolo-X.tflite](https://huggingface.co/qualcomm/Yolo-X/blob/main/Yolo-X.tflite) | | Yolo-X | float | SA8650 (Proxy) | Qualcomm® SA8650P (Proxy) | QNN | 15.705 ms | 5 - 8 MB | NPU | Use Export Script | | Yolo-X | float | SA8775P ADP | Qualcomm® SA8775P | TFLITE | 26.709 ms | 0 - 76 MB | NPU | [Yolo-X.tflite](https://huggingface.co/qualcomm/Yolo-X/blob/main/Yolo-X.tflite) | | Yolo-X | float | SA8775P ADP | Qualcomm® SA8775P | QNN | 22.043 ms | 0 - 13 MB | NPU | Use Export Script | | Yolo-X | float | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 Mobile | TFLITE | 20.57 ms | 0 - 16 MB | NPU | [Yolo-X.tflite](https://huggingface.co/qualcomm/Yolo-X/blob/main/Yolo-X.tflite) | | Yolo-X | float | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 Mobile | QNN | 15.753 ms | 5 - 22 MB | NPU | Use Export Script | | Yolo-X | float | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 Mobile | ONNX | 20.337 ms | 1 - 151 MB | NPU | [Yolo-X.onnx](https://huggingface.co/qualcomm/Yolo-X/blob/main/Yolo-X.onnx) | | Yolo-X | float | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | TFLITE | 15.229 ms | 0 - 111 MB | NPU | [Yolo-X.tflite](https://huggingface.co/qualcomm/Yolo-X/blob/main/Yolo-X.tflite) | | Yolo-X | float | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | QNN | 12.042 ms | 5 - 54 MB | NPU | Use Export Script | | Yolo-X | float | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | ONNX | 15.441 ms | 5 - 69 MB | NPU | [Yolo-X.onnx](https://huggingface.co/qualcomm/Yolo-X/blob/main/Yolo-X.onnx) | | Yolo-X | float | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite Mobile | TFLITE | 13.711 ms | 0 - 79 MB | NPU | [Yolo-X.tflite](https://huggingface.co/qualcomm/Yolo-X/blob/main/Yolo-X.tflite) | | Yolo-X | float | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite Mobile | QNN | 9.209 ms | 5 - 51 MB | NPU | Use Export Script | | Yolo-X | float | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite Mobile | ONNX | 14.343 ms | 5 - 59 MB | NPU | [Yolo-X.onnx](https://huggingface.co/qualcomm/Yolo-X/blob/main/Yolo-X.onnx) | | Yolo-X | float | Snapdragon X Elite CRD | Snapdragon® X Elite | QNN | 15.95 ms | 5 - 5 MB | NPU | Use Export Script | | Yolo-X | float | Snapdragon X Elite CRD | Snapdragon® X Elite | ONNX | 20.752 ms | 46 - 46 MB | NPU | [Yolo-X.onnx](https://huggingface.co/qualcomm/Yolo-X/blob/main/Yolo-X.onnx) | | Yolo-X | w8a16 | QCS8275 (Proxy) | Qualcomm® QCS8275 (Proxy) | QNN | 24.942 ms | 2 - 12 MB | NPU | Use Export Script | | Yolo-X | w8a16 | QCS8450 (Proxy) | Qualcomm® QCS8450 (Proxy) | QNN | 15.586 ms | 2 - 84 MB | NPU | Use Export Script | | Yolo-X | w8a16 | QCS8550 (Proxy) | Qualcomm® QCS8550 (Proxy) | QNN | 10.29 ms | 2 - 6 MB | NPU | Use Export Script | | Yolo-X | w8a16 | QCS9075 (Proxy) | Qualcomm® QCS9075 (Proxy) | QNN | 10.703 ms | 1 - 16 MB | NPU | Use Export Script | | Yolo-X | w8a16 | RB3 Gen 2 (Proxy) | Qualcomm® QCS6490 (Proxy) | QNN | 53.241 ms | 2 - 16 MB | NPU | Use Export Script | | Yolo-X | w8a16 | SA7255P ADP | Qualcomm® SA7255P | QNN | 24.942 ms | 2 - 12 MB | NPU | Use Export Script | | Yolo-X | w8a16 | SA8255 (Proxy) | Qualcomm® SA8255P (Proxy) | QNN | 10.449 ms | 2 - 5 MB | NPU | Use Export Script | | Yolo-X | w8a16 | SA8295P ADP | Qualcomm® SA8295P | QNN | 14.409 ms | 0 - 17 MB | NPU | Use Export Script | | Yolo-X | w8a16 | SA8650 (Proxy) | Qualcomm® SA8650P (Proxy) | QNN | 10.548 ms | 3 - 5 MB | NPU | Use Export Script | | Yolo-X | w8a16 | SA8775P ADP | Qualcomm® SA8775P | QNN | 10.703 ms | 1 - 16 MB | NPU | Use Export Script | | Yolo-X | w8a16 | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 Mobile | QNN | 10.461 ms | 2 - 17 MB | NPU | Use Export Script | | Yolo-X | w8a16 | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 Mobile | ONNX | 18.693 ms | 1 - 48 MB | NPU | [Yolo-X.onnx](https://huggingface.co/qualcomm/Yolo-X/blob/main/Yolo-X_w8a16.onnx) | | Yolo-X | w8a16 | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | QNN | 6.951 ms | 2 - 80 MB | NPU | Use Export Script | | Yolo-X | w8a16 | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | ONNX | 13.603 ms | 2 - 103 MB | NPU | [Yolo-X.onnx](https://huggingface.co/qualcomm/Yolo-X/blob/main/Yolo-X_w8a16.onnx) | | Yolo-X | w8a16 | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite Mobile | QNN | 5.952 ms | 2 - 73 MB | NPU | Use Export Script | | Yolo-X | w8a16 | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite Mobile | ONNX | 14.152 ms | 4 - 90 MB | NPU | [Yolo-X.onnx](https://huggingface.co/qualcomm/Yolo-X/blob/main/Yolo-X_w8a16.onnx) | | Yolo-X | w8a16 | Snapdragon X Elite CRD | Snapdragon® X Elite | QNN | 11.179 ms | 2 - 2 MB | NPU | Use Export Script | | Yolo-X | w8a16 | Snapdragon X Elite CRD | Snapdragon® X Elite | ONNX | 20.376 ms | 24 - 24 MB | NPU | [Yolo-X.onnx](https://huggingface.co/qualcomm/Yolo-X/blob/main/Yolo-X_w8a16.onnx) | ## Installation Install the package via pip: ```bash pip install "qai-hub-models[yolox]" ``` ## 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.yolox.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.yolox.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.yolox.export ``` ``` Profiling Results ------------------------------------------------------------ Yolo-X Device : cs_8275 (ANDROID 14) Runtime : TFLITE Estimated inference time (ms) : 79.8 Estimated peak memory usage (MB): [0, 75] Total # Ops : 418 Compute Unit(s) : npu (418 ops) gpu (0 ops) cpu (0 ops) ``` ## How does this work? This [export script](https://aihub.qualcomm.com/models/yolox/qai_hub_models/models/Yolo-X/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.yolox 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.yolox.demo --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.yolox.demo -- --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-X's performance across various devices [here](https://aihub.qualcomm.com/models/yolox). Explore all available models on [Qualcomm® AI Hub](https://aihub.qualcomm.com/) ## License * The license for the original implementation of Yolo-X can be found [here](https://github.com/Megvii-BaseDetection/YOLOX/blob/main/LICENSE). * The license for the compiled assets for on-device deployment can be found [here](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/Qualcomm+AI+Hub+Proprietary+License.pdf) ## References * [YOLOX: Exceeding YOLO Series in 2021](https://github.com/Megvii-BaseDetection/YOLOX/blob/main/README.md) * [Source Model Implementation](https://github.com/Megvii-BaseDetection/YOLOX/) ## 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).