Yolo-v6: Optimized for Mobile Deployment

Real-time object detection optimized for mobile and edge

YoloV6 is a machine learning model that predicts bounding boxes and classes of objects in an image.

This model is an implementation of Yolo-v6 found here.

This repository provides scripts to run Yolo-v6 on Qualcomm® devices. More details on model performance across various devices, can be found here.

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: YoloV6-N
- Input resolution: 640x640
- Number of parameters: 4.68M
- Model size (float): 17.9 MB
- Model size (w8a8): 4.68 MB
- Model size (w8a16): 5.03 MB

Model	Precision	Device	Chipset	Target Runtime	Inference Time (ms)	Peak Memory Range (MB)	Primary Compute Unit	Target Model
Yolo-v6	float	QCS8275 (Proxy)	Qualcomm® QCS8275 (Proxy)	TFLITE	17.711 ms	0 - 51 MB	NPU	--
Yolo-v6	float	QCS8275 (Proxy)	Qualcomm® QCS8275 (Proxy)	QNN_DLC	15.013 ms	3 - 73 MB	NPU	--
Yolo-v6	float	QCS8450 (Proxy)	Qualcomm® QCS8450 (Proxy)	TFLITE	7.236 ms	0 - 44 MB	NPU	--
Yolo-v6	float	QCS8450 (Proxy)	Qualcomm® QCS8450 (Proxy)	QNN_DLC	7.678 ms	5 - 39 MB	NPU	--
Yolo-v6	float	QCS8550 (Proxy)	Qualcomm® QCS8550 (Proxy)	TFLITE	6.292 ms	0 - 16 MB	NPU	--
Yolo-v6	float	QCS8550 (Proxy)	Qualcomm® QCS8550 (Proxy)	QNN_DLC	4.745 ms	0 - 38 MB	NPU	--
Yolo-v6	float	QCS9075 (Proxy)	Qualcomm® QCS9075 (Proxy)	TFLITE	7.914 ms	0 - 52 MB	NPU	--
Yolo-v6	float	QCS9075 (Proxy)	Qualcomm® QCS9075 (Proxy)	QNN_DLC	6.491 ms	0 - 69 MB	NPU	--
Yolo-v6	float	SA7255P ADP	Qualcomm® SA7255P	TFLITE	17.711 ms	0 - 51 MB	NPU	--
Yolo-v6	float	SA7255P ADP	Qualcomm® SA7255P	QNN_DLC	15.013 ms	3 - 73 MB	NPU	--
Yolo-v6	float	SA8255 (Proxy)	Qualcomm® SA8255P (Proxy)	TFLITE	6.098 ms	0 - 15 MB	NPU	--
Yolo-v6	float	SA8255 (Proxy)	Qualcomm® SA8255P (Proxy)	QNN_DLC	4.761 ms	0 - 38 MB	NPU	--
Yolo-v6	float	SA8295P ADP	Qualcomm® SA8295P	TFLITE	8.635 ms	0 - 36 MB	NPU	--
Yolo-v6	float	SA8295P ADP	Qualcomm® SA8295P	QNN_DLC	7.361 ms	3 - 35 MB	NPU	--
Yolo-v6	float	SA8650 (Proxy)	Qualcomm® SA8650P (Proxy)	TFLITE	6.126 ms	0 - 15 MB	NPU	--
Yolo-v6	float	SA8650 (Proxy)	Qualcomm® SA8650P (Proxy)	QNN_DLC	4.766 ms	0 - 38 MB	NPU	--
Yolo-v6	float	SA8775P ADP	Qualcomm® SA8775P	TFLITE	7.914 ms	0 - 52 MB	NPU	--
Yolo-v6	float	SA8775P ADP	Qualcomm® SA8775P	QNN_DLC	6.491 ms	0 - 69 MB	NPU	--
Yolo-v6	float	Samsung Galaxy S23	Snapdragon® 8 Gen 2 Mobile	TFLITE	6.096 ms	0 - 21 MB	NPU	--
Yolo-v6	float	Samsung Galaxy S23	Snapdragon® 8 Gen 2 Mobile	QNN_DLC	4.749 ms	0 - 38 MB	NPU	--
Yolo-v6	float	Samsung Galaxy S23	Snapdragon® 8 Gen 2 Mobile	ONNX	5.942 ms	0 - 51 MB	NPU	--
Yolo-v6	float	Samsung Galaxy S24	Snapdragon® 8 Gen 3 Mobile	TFLITE	4.25 ms	0 - 62 MB	NPU	--
Yolo-v6	float	Samsung Galaxy S24	Snapdragon® 8 Gen 3 Mobile	QNN_DLC	3.528 ms	5 - 121 MB	NPU	--
Yolo-v6	float	Samsung Galaxy S24	Snapdragon® 8 Gen 3 Mobile	ONNX	4.238 ms	5 - 152 MB	NPU	--
Yolo-v6	float	Snapdragon 8 Elite QRD	Snapdragon® 8 Elite Mobile	TFLITE	3.567 ms	0 - 57 MB	NPU	--
Yolo-v6	float	Snapdragon 8 Elite QRD	Snapdragon® 8 Elite Mobile	QNN_DLC	3.122 ms	5 - 77 MB	NPU	--
Yolo-v6	float	Snapdragon 8 Elite QRD	Snapdragon® 8 Elite Mobile	ONNX	4.161 ms	5 - 94 MB	NPU	--
Yolo-v6	float	Snapdragon X Elite CRD	Snapdragon® X Elite	QNN_DLC	5.061 ms	31 - 31 MB	NPU	--
Yolo-v6	float	Snapdragon X Elite CRD	Snapdragon® X Elite	ONNX	6.068 ms	7 - 7 MB	NPU	--
Yolo-v6	w8a16	QCS8275 (Proxy)	Qualcomm® QCS8275 (Proxy)	QNN_DLC	5.162 ms	2 - 33 MB	NPU	--
Yolo-v6	w8a16	QCS8450 (Proxy)	Qualcomm® QCS8450 (Proxy)	QNN_DLC	2.645 ms	2 - 41 MB	NPU	--
Yolo-v6	w8a16	QCS8550 (Proxy)	Qualcomm® QCS8550 (Proxy)	QNN_DLC	2.125 ms	2 - 12 MB	NPU	--
Yolo-v6	w8a16	QCS9075 (Proxy)	Qualcomm® QCS9075 (Proxy)	QNN_DLC	2.744 ms	1 - 28 MB	NPU	--
Yolo-v6	w8a16	RB3 Gen 2 (Proxy)	Qualcomm® QCS6490 (Proxy)	QNN_DLC	8.788 ms	2 - 34 MB	NPU	--
Yolo-v6	w8a16	SA7255P ADP	Qualcomm® SA7255P	QNN_DLC	5.162 ms	2 - 33 MB	NPU	--
Yolo-v6	w8a16	SA8255 (Proxy)	Qualcomm® SA8255P (Proxy)	QNN_DLC	2.131 ms	2 - 13 MB	NPU	--
Yolo-v6	w8a16	SA8295P ADP	Qualcomm® SA8295P	QNN_DLC	3.295 ms	2 - 40 MB	NPU	--
Yolo-v6	w8a16	SA8650 (Proxy)	Qualcomm® SA8650P (Proxy)	QNN_DLC	2.139 ms	1 - 11 MB	NPU	--
Yolo-v6	w8a16	SA8775P ADP	Qualcomm® SA8775P	QNN_DLC	2.744 ms	1 - 28 MB	NPU	--
Yolo-v6	w8a16	Samsung Galaxy S23	Snapdragon® 8 Gen 2 Mobile	QNN_DLC	2.13 ms	2 - 13 MB	NPU	--
Yolo-v6	w8a16	Samsung Galaxy S23	Snapdragon® 8 Gen 2 Mobile	ONNX	43.981 ms	36 - 154 MB	NPU	--
Yolo-v6	w8a16	Samsung Galaxy S24	Snapdragon® 8 Gen 3 Mobile	QNN_DLC	1.436 ms	2 - 46 MB	NPU	--
Yolo-v6	w8a16	Samsung Galaxy S24	Snapdragon® 8 Gen 3 Mobile	ONNX	33.71 ms	18 - 846 MB	NPU	--
Yolo-v6	w8a16	Snapdragon 8 Elite QRD	Snapdragon® 8 Elite Mobile	QNN_DLC	1.321 ms	2 - 33 MB	NPU	--
Yolo-v6	w8a16	Snapdragon 8 Elite QRD	Snapdragon® 8 Elite Mobile	ONNX	32.159 ms	30 - 996 MB	NPU	--
Yolo-v6	w8a16	Snapdragon X Elite CRD	Snapdragon® X Elite	QNN_DLC	2.409 ms	0 - 0 MB	NPU	--
Yolo-v6	w8a16	Snapdragon X Elite CRD	Snapdragon® X Elite	ONNX	46.172 ms	58 - 58 MB	NPU	--
Yolo-v6	w8a8	QCS8275 (Proxy)	Qualcomm® QCS8275 (Proxy)	TFLITE	4.478 ms	0 - 25 MB	NPU	--
Yolo-v6	w8a8	QCS8275 (Proxy)	Qualcomm® QCS8275 (Proxy)	QNN_DLC	6.532 ms	1 - 27 MB	NPU	--
Yolo-v6	w8a8	QCS8450 (Proxy)	Qualcomm® QCS8450 (Proxy)	TFLITE	2.261 ms	0 - 41 MB	NPU	--
Yolo-v6	w8a8	QCS8450 (Proxy)	Qualcomm® QCS8450 (Proxy)	QNN_DLC	1.693 ms	0 - 38 MB	NPU	--
Yolo-v6	w8a8	QCS8550 (Proxy)	Qualcomm® QCS8550 (Proxy)	TFLITE	2.095 ms	0 - 29 MB	NPU	--
Yolo-v6	w8a8	QCS8550 (Proxy)	Qualcomm® QCS8550 (Proxy)	QNN_DLC	3.04 ms	1 - 28 MB	NPU	--
Yolo-v6	w8a8	QCS9075 (Proxy)	Qualcomm® QCS9075 (Proxy)	TFLITE	2.579 ms	0 - 25 MB	NPU	--
Yolo-v6	w8a8	QCS9075 (Proxy)	Qualcomm® QCS9075 (Proxy)	QNN_DLC	3.451 ms	1 - 27 MB	NPU	--
Yolo-v6	w8a8	RB3 Gen 2 (Proxy)	Qualcomm® QCS6490 (Proxy)	TFLITE	4.528 ms	0 - 32 MB	NPU	--
Yolo-v6	w8a8	RB3 Gen 2 (Proxy)	Qualcomm® QCS6490 (Proxy)	QNN_DLC	5.598 ms	1 - 32 MB	NPU	--
Yolo-v6	w8a8	SA7255P ADP	Qualcomm® SA7255P	TFLITE	4.478 ms	0 - 25 MB	NPU	--
Yolo-v6	w8a8	SA7255P ADP	Qualcomm® SA7255P	QNN_DLC	6.532 ms	1 - 27 MB	NPU	--
Yolo-v6	w8a8	SA8255 (Proxy)	Qualcomm® SA8255P (Proxy)	TFLITE	2.107 ms	0 - 30 MB	NPU	--
Yolo-v6	w8a8	SA8255 (Proxy)	Qualcomm® SA8255P (Proxy)	QNN_DLC	3.054 ms	1 - 28 MB	NPU	--
Yolo-v6	w8a8	SA8295P ADP	Qualcomm® SA8295P	TFLITE	2.967 ms	0 - 33 MB	NPU	--
Yolo-v6	w8a8	SA8295P ADP	Qualcomm® SA8295P	QNN_DLC	2.271 ms	1 - 34 MB	NPU	--
Yolo-v6	w8a8	SA8650 (Proxy)	Qualcomm® SA8650P (Proxy)	TFLITE	2.105 ms	0 - 29 MB	NPU	--
Yolo-v6	w8a8	SA8650 (Proxy)	Qualcomm® SA8650P (Proxy)	QNN_DLC	3.048 ms	1 - 28 MB	NPU	--
Yolo-v6	w8a8	SA8775P ADP	Qualcomm® SA8775P	TFLITE	2.579 ms	0 - 25 MB	NPU	--
Yolo-v6	w8a8	SA8775P ADP	Qualcomm® SA8775P	QNN_DLC	3.451 ms	1 - 27 MB	NPU	--
Yolo-v6	w8a8	Samsung Galaxy S23	Snapdragon® 8 Gen 2 Mobile	TFLITE	2.107 ms	0 - 30 MB	NPU	--
Yolo-v6	w8a8	Samsung Galaxy S23	Snapdragon® 8 Gen 2 Mobile	QNN_DLC	3.052 ms	1 - 28 MB	NPU	--
Yolo-v6	w8a8	Samsung Galaxy S23	Snapdragon® 8 Gen 2 Mobile	ONNX	37.17 ms	24 - 146 MB	NPU	--
Yolo-v6	w8a8	Samsung Galaxy S24	Snapdragon® 8 Gen 3 Mobile	TFLITE	1.368 ms	0 - 44 MB	NPU	--
Yolo-v6	w8a8	Samsung Galaxy S24	Snapdragon® 8 Gen 3 Mobile	QNN_DLC	2.195 ms	1 - 32 MB	NPU	--
Yolo-v6	w8a8	Samsung Galaxy S24	Snapdragon® 8 Gen 3 Mobile	ONNX	30.685 ms	25 - 761 MB	NPU	--
Yolo-v6	w8a8	Snapdragon 8 Elite QRD	Snapdragon® 8 Elite Mobile	TFLITE	1.369 ms	0 - 27 MB	NPU	--
Yolo-v6	w8a8	Snapdragon 8 Elite QRD	Snapdragon® 8 Elite Mobile	QNN_DLC	2.305 ms	1 - 33 MB	NPU	--
Yolo-v6	w8a8	Snapdragon 8 Elite QRD	Snapdragon® 8 Elite Mobile	ONNX	30.592 ms	27 - 879 MB	NPU	--
Yolo-v6	w8a8	Snapdragon X Elite CRD	Snapdragon® X Elite	QNN_DLC	3.273 ms	13 - 13 MB	NPU	--
Yolo-v6	w8a8	Snapdragon X Elite CRD	Snapdragon® X Elite	ONNX	40.672 ms	55 - 55 MB	NPU	--

Installation

Install the package via pip:

pip install "qai-hub-models[yolov6]"

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.yolov6.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.yolov6.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.yolov6.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.yolov6 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.yolov6.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.yolov6.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-v6's performance across various devices here. Explore all available models on Qualcomm® AI Hub

License

The license for the original implementation of Yolo-v6 can be found here.
The license for the compiled assets for on-device deployment can be found here

References

Community

Join our AI Hub Slack community to collaborate, post questions and learn more about on-device AI.
For questions or feedback please reach out to us.