object_detection_nanodet

Nanodet

Nanodet: NanoDet is a FCOS-style one-stage anchor-free object detection model which using Generalized Focal Loss as classification and regression loss.In NanoDet-Plus, we propose a novel label assignment strategy with a simple assign guidance module (AGM) and a dynamic soft label assigner (DSLA) to solve the optimal label assignment problem in lightweight model training.

Note:

• This version of nanodet: Nanodet-m-plus-1.5x_416
• object_detection_nanodet_2022nov_int8bq.onnx represents the block-quantized version in int8 precision and is generated using block_quantize.py with block_size=64.

Demo

Python

Run the following command to try the demo:

# detect on camera input
python demo.py
# detect on an image
python demo.py --input /path/to/image -v

Note:

• image result saved as "result.jpg"

C++

Install latest OpenCV and CMake >= 3.24.0 to get started with:

# A typical and default installation path of OpenCV is /usr/local
cmake -B build -D OPENCV_INSTALLATION_PATH=/path/to/opencv/installation .
cmake --build build

# detect on camera input
./build/opencv_zoo_object_detection_nanodet
# detect on an image
./build/opencv_zoo_object_detection_nanodet -i=/path/to/image
# get help messages
./build/opencv_zoo_object_detection_nanodet -h

Results

Here are some of the sample results that were observed using the model,

Check benchmark/download_data.py for the original images.

Video inference result,

Model metrics:

The model is evaluated on COCO 2017 val. Results are showed below:

<table> <tr><th>Average Precision </th><th>Average Recall</th></tr> <tr><td> </td><td> </td></tr> </table>

License

All files in this directory are licensed under Apache 2.0 License.

Contributor Details

• Google Summer of Code'22
• Contributor: Sri Siddarth Chakaravarthy
• Github Profile: https://github.com/Sidd1609
• Organisation: OpenCV
• Project: Lightweight object detection models using OpenCV

Reference

• Nanodet: https://zhuanlan.zhihu.com/p/306530300
• Nanodet Plus: https://zhuanlan.zhihu.com/p/449912627
• Nanodet weight and scripts for training: https://github.com/RangiLyu/nanodet