Hugging Face's logo

qualcomm
/
Yolo-X

Object Detection
PyTorch
LiteRT
real_time
android
Model card Files Community

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.

This repository provides scripts to run Yolo-X on Qualcomm® devices. More details on model performance across various devices, can be found here.

Model Details

  • Model Type: Model_use_case.object_detection
  • Model Stats:
    • Model checkpoint: YoloX Small
    • Input resolution: 640x640
    • Number of parameters: 8.98M
    • Model size (float): 34.3 MB
    • Model size (w8a16): 9.53 MB
    • Model size (w8a8): 8.96 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 37.192 ms 0 - 34 MB NPU Yolo-X.tflite
Yolo-X float QCS8275 (Proxy) Qualcomm® QCS8275 (Proxy) QNN_DLC 31.392 ms 4 - 61 MB NPU Yolo-X.dlc
Yolo-X float QCS8450 (Proxy) Qualcomm® QCS8450 (Proxy) TFLITE 19.743 ms 0 - 51 MB NPU Yolo-X.tflite
Yolo-X float QCS8450 (Proxy) Qualcomm® QCS8450 (Proxy) QNN_DLC 17.939 ms 5 - 47 MB NPU Yolo-X.dlc
Yolo-X float QCS8550 (Proxy) Qualcomm® QCS8550 (Proxy) TFLITE 14.098 ms 0 - 9 MB NPU Yolo-X.tflite
Yolo-X float QCS8550 (Proxy) Qualcomm® QCS8550 (Proxy) QNN_DLC 8.247 ms 5 - 24 MB NPU Yolo-X.dlc
Yolo-X float QCS9075 (Proxy) Qualcomm® QCS9075 (Proxy) TFLITE 16.975 ms 0 - 35 MB NPU Yolo-X.tflite
Yolo-X float QCS9075 (Proxy) Qualcomm® QCS9075 (Proxy) QNN_DLC 11.293 ms 0 - 59 MB NPU Yolo-X.dlc
Yolo-X float SA7255P ADP Qualcomm® SA7255P TFLITE 37.192 ms 0 - 34 MB NPU Yolo-X.tflite
Yolo-X float SA7255P ADP Qualcomm® SA7255P QNN_DLC 31.392 ms 4 - 61 MB NPU Yolo-X.dlc
Yolo-X float SA8255 (Proxy) Qualcomm® SA8255P (Proxy) TFLITE 14.332 ms 0 - 10 MB NPU Yolo-X.tflite
Yolo-X float SA8255 (Proxy) Qualcomm® SA8255P (Proxy) QNN_DLC 8.267 ms 5 - 20 MB NPU Yolo-X.dlc
Yolo-X float SA8295P ADP Qualcomm® SA8295P TFLITE 21.877 ms 0 - 40 MB NPU Yolo-X.tflite
Yolo-X float SA8295P ADP Qualcomm® SA8295P QNN_DLC 14.898 ms 2 - 38 MB NPU Yolo-X.dlc
Yolo-X float SA8650 (Proxy) Qualcomm® SA8650P (Proxy) TFLITE 13.909 ms 0 - 9 MB NPU Yolo-X.tflite
Yolo-X float SA8650 (Proxy) Qualcomm® SA8650P (Proxy) QNN_DLC 8.267 ms 5 - 20 MB NPU Yolo-X.dlc
Yolo-X float SA8775P ADP Qualcomm® SA8775P TFLITE 16.975 ms 0 - 35 MB NPU Yolo-X.tflite
Yolo-X float SA8775P ADP Qualcomm® SA8775P QNN_DLC 11.293 ms 0 - 59 MB NPU Yolo-X.dlc
Yolo-X float Samsung Galaxy S23 Snapdragon® 8 Gen 2 Mobile TFLITE 14.388 ms 0 - 10 MB NPU Yolo-X.tflite
Yolo-X float Samsung Galaxy S23 Snapdragon® 8 Gen 2 Mobile QNN_DLC 8.28 ms 5 - 17 MB NPU Yolo-X.dlc
Yolo-X float Samsung Galaxy S23 Snapdragon® 8 Gen 2 Mobile ONNX 12.956 ms 0 - 63 MB NPU Yolo-X.onnx
Yolo-X float Samsung Galaxy S24 Snapdragon® 8 Gen 3 Mobile TFLITE 9.888 ms 0 - 47 MB NPU Yolo-X.tflite
Yolo-X float Samsung Galaxy S24 Snapdragon® 8 Gen 3 Mobile QNN_DLC 5.99 ms 5 - 75 MB NPU Yolo-X.dlc
Yolo-X float Samsung Galaxy S24 Snapdragon® 8 Gen 3 Mobile ONNX 10.039 ms 5 - 150 MB NPU Yolo-X.onnx
Yolo-X float Snapdragon 8 Elite QRD Snapdragon® 8 Elite Mobile TFLITE 9.1 ms 0 - 41 MB NPU Yolo-X.tflite
Yolo-X float Snapdragon 8 Elite QRD Snapdragon® 8 Elite Mobile QNN_DLC 5.722 ms 5 - 77 MB NPU Yolo-X.dlc
Yolo-X float Snapdragon 8 Elite QRD Snapdragon® 8 Elite Mobile ONNX 8.187 ms 5 - 104 MB NPU Yolo-X.onnx
Yolo-X float Snapdragon X Elite CRD Snapdragon® X Elite QNN_DLC 9.596 ms 2 - 2 MB NPU Yolo-X.dlc
Yolo-X float Snapdragon X Elite CRD Snapdragon® X Elite ONNX 13.307 ms 15 - 15 MB NPU Yolo-X.onnx
Yolo-X w8a16 QCS8275 (Proxy) Qualcomm® QCS8275 (Proxy) QNN_DLC 15.638 ms 2 - 37 MB NPU Yolo-X.dlc
Yolo-X w8a16 QCS8450 (Proxy) Qualcomm® QCS8450 (Proxy) QNN_DLC 9.151 ms 2 - 57 MB NPU Yolo-X.dlc
Yolo-X w8a16 QCS8550 (Proxy) Qualcomm® QCS8550 (Proxy) QNN_DLC 7.504 ms 2 - 14 MB NPU Yolo-X.dlc
Yolo-X w8a16 QCS9075 (Proxy) Qualcomm® QCS9075 (Proxy) QNN_DLC 8.401 ms 1 - 36 MB NPU Yolo-X.dlc
Yolo-X w8a16 RB3 Gen 2 (Proxy) Qualcomm® QCS6490 (Proxy) QNN_DLC 26.587 ms 0 - 41 MB NPU Yolo-X.dlc
Yolo-X w8a16 SA7255P ADP Qualcomm® SA7255P QNN_DLC 15.638 ms 2 - 37 MB NPU Yolo-X.dlc
Yolo-X w8a16 SA8255 (Proxy) Qualcomm® SA8255P (Proxy) QNN_DLC 7.503 ms 0 - 16 MB NPU Yolo-X.dlc
Yolo-X w8a16 SA8295P ADP Qualcomm® SA8295P QNN_DLC 9.944 ms 2 - 40 MB NPU Yolo-X.dlc
Yolo-X w8a16 SA8650 (Proxy) Qualcomm® SA8650P (Proxy) QNN_DLC 7.501 ms 2 - 14 MB NPU Yolo-X.dlc
Yolo-X w8a16 SA8775P ADP Qualcomm® SA8775P QNN_DLC 8.401 ms 1 - 36 MB NPU Yolo-X.dlc
Yolo-X w8a16 Samsung Galaxy S23 Snapdragon® 8 Gen 2 Mobile QNN_DLC 7.5 ms 2 - 15 MB NPU Yolo-X.dlc
Yolo-X w8a16 Samsung Galaxy S23 Snapdragon® 8 Gen 2 Mobile ONNX 14.218 ms 0 - 51 MB NPU Yolo-X.onnx
Yolo-X w8a16 Samsung Galaxy S24 Snapdragon® 8 Gen 3 Mobile QNN_DLC 5.075 ms 2 - 50 MB NPU Yolo-X.dlc
Yolo-X w8a16 Samsung Galaxy S24 Snapdragon® 8 Gen 3 Mobile ONNX 10.418 ms 2 - 92 MB NPU Yolo-X.onnx
Yolo-X w8a16 Snapdragon 8 Elite QRD Snapdragon® 8 Elite Mobile QNN_DLC 4.813 ms 2 - 52 MB NPU Yolo-X.dlc
Yolo-X w8a16 Snapdragon 8 Elite QRD Snapdragon® 8 Elite Mobile ONNX 11.801 ms 2 - 89 MB NPU Yolo-X.onnx
Yolo-X w8a16 Snapdragon X Elite CRD Snapdragon® X Elite QNN_DLC 9.081 ms 0 - 0 MB NPU Yolo-X.dlc
Yolo-X w8a16 Snapdragon X Elite CRD Snapdragon® X Elite ONNX 15.278 ms 7 - 7 MB NPU Yolo-X.onnx
Yolo-X w8a8 QCS8275 (Proxy) Qualcomm® QCS8275 (Proxy) TFLITE 6.4 ms 0 - 28 MB NPU Yolo-X.tflite
Yolo-X w8a8 QCS8275 (Proxy) Qualcomm® QCS8275 (Proxy) QNN_DLC 5.465 ms 1 - 33 MB NPU Yolo-X.dlc
Yolo-X w8a8 QCS8450 (Proxy) Qualcomm® QCS8450 (Proxy) TFLITE 3.18 ms 0 - 44 MB NPU Yolo-X.tflite
Yolo-X w8a8 QCS8450 (Proxy) Qualcomm® QCS8450 (Proxy) QNN_DLC 2.85 ms 1 - 51 MB NPU Yolo-X.dlc
Yolo-X w8a8 QCS8550 (Proxy) Qualcomm® QCS8550 (Proxy) TFLITE 2.879 ms 0 - 35 MB NPU Yolo-X.tflite
Yolo-X w8a8 QCS8550 (Proxy) Qualcomm® QCS8550 (Proxy) QNN_DLC 2.302 ms 1 - 13 MB NPU Yolo-X.dlc
Yolo-X w8a8 QCS9075 (Proxy) Qualcomm® QCS9075 (Proxy) TFLITE 3.264 ms 0 - 29 MB NPU Yolo-X.tflite
Yolo-X w8a8 QCS9075 (Proxy) Qualcomm® QCS9075 (Proxy) QNN_DLC 2.664 ms 1 - 34 MB NPU Yolo-X.dlc
Yolo-X w8a8 RB3 Gen 2 (Proxy) Qualcomm® QCS6490 (Proxy) TFLITE 8.202 ms 0 - 39 MB NPU Yolo-X.tflite
Yolo-X w8a8 RB3 Gen 2 (Proxy) Qualcomm® QCS6490 (Proxy) QNN_DLC 9.708 ms 1 - 40 MB NPU Yolo-X.dlc
Yolo-X w8a8 SA7255P ADP Qualcomm® SA7255P TFLITE 6.4 ms 0 - 28 MB NPU Yolo-X.tflite
Yolo-X w8a8 SA7255P ADP Qualcomm® SA7255P QNN_DLC 5.465 ms 1 - 33 MB NPU Yolo-X.dlc
Yolo-X w8a8 SA8255 (Proxy) Qualcomm® SA8255P (Proxy) TFLITE 2.88 ms 0 - 34 MB NPU Yolo-X.tflite
Yolo-X w8a8 SA8255 (Proxy) Qualcomm® SA8255P (Proxy) QNN_DLC 2.291 ms 1 - 12 MB NPU Yolo-X.dlc
Yolo-X w8a8 SA8295P ADP Qualcomm® SA8295P TFLITE 4.17 ms 0 - 31 MB NPU Yolo-X.tflite
Yolo-X w8a8 SA8295P ADP Qualcomm® SA8295P QNN_DLC 3.585 ms 1 - 35 MB NPU Yolo-X.dlc
Yolo-X w8a8 SA8650 (Proxy) Qualcomm® SA8650P (Proxy) TFLITE 2.879 ms 0 - 10 MB NPU Yolo-X.tflite
Yolo-X w8a8 SA8650 (Proxy) Qualcomm® SA8650P (Proxy) QNN_DLC 2.293 ms 2 - 13 MB NPU Yolo-X.dlc
Yolo-X w8a8 SA8775P ADP Qualcomm® SA8775P TFLITE 3.264 ms 0 - 29 MB NPU Yolo-X.tflite
Yolo-X w8a8 SA8775P ADP Qualcomm® SA8775P QNN_DLC 2.664 ms 1 - 34 MB NPU Yolo-X.dlc
Yolo-X w8a8 Samsung Galaxy S23 Snapdragon® 8 Gen 2 Mobile TFLITE 2.87 ms 0 - 34 MB NPU Yolo-X.tflite
Yolo-X w8a8 Samsung Galaxy S23 Snapdragon® 8 Gen 2 Mobile QNN_DLC 2.302 ms 1 - 12 MB NPU Yolo-X.dlc
Yolo-X w8a8 Samsung Galaxy S23 Snapdragon® 8 Gen 2 Mobile ONNX 9.647 ms 0 - 48 MB NPU Yolo-X.onnx
Yolo-X w8a8 Samsung Galaxy S24 Snapdragon® 8 Gen 3 Mobile TFLITE 1.89 ms 0 - 48 MB NPU Yolo-X.tflite
Yolo-X w8a8 Samsung Galaxy S24 Snapdragon® 8 Gen 3 Mobile QNN_DLC 1.556 ms 1 - 52 MB NPU Yolo-X.dlc
Yolo-X w8a8 Samsung Galaxy S24 Snapdragon® 8 Gen 3 Mobile ONNX 6.995 ms 1 - 118 MB NPU Yolo-X.onnx
Yolo-X w8a8 Snapdragon 8 Elite QRD Snapdragon® 8 Elite Mobile TFLITE 1.622 ms 0 - 34 MB NPU Yolo-X.tflite
Yolo-X w8a8 Snapdragon 8 Elite QRD Snapdragon® 8 Elite Mobile QNN_DLC 1.301 ms 1 - 39 MB NPU Yolo-X.dlc
Yolo-X w8a8 Snapdragon 8 Elite QRD Snapdragon® 8 Elite Mobile ONNX 7.879 ms 1 - 74 MB NPU Yolo-X.onnx
Yolo-X w8a8 Snapdragon X Elite CRD Snapdragon® X Elite QNN_DLC 2.862 ms 24 - 24 MB NPU Yolo-X.dlc
Yolo-X w8a8 Snapdragon X Elite CRD Snapdragon® X Elite ONNX 10.042 ms 8 - 8 MB NPU Yolo-X.onnx

Installation

Install the package via pip:

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 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.

qai-hub configure --api_token API_TOKEN

Navigate to 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.

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.
python -m qai_hub_models.models.yolox.export

How does this work?

This export script leverages Qualcomm® AI Hub 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.

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.

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.

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.

Run demo on a cloud-hosted device

You can also run the demo on-device.

python -m qai_hub_models.models.yolox.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.yolox.demo -- --eval-mode on-device

Deploying compiled model to Android

The models can be deployed using multiple runtimes:

  • TensorFlow Lite (.tflite export): This tutorial provides a guide to deploy the .tflite model in an Android application.

  • QNN (.so export ): This sample app 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. Explore all available models on Qualcomm® AI Hub

License

  • The license for the original implementation of Yolo-X can be found here.
  • The license for the compiled assets for on-device deployment can be found here

References

Community

Downloads last month
482
Inference Providers NEW
Object Detection
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support