PP-PicoDet

PicoDet is an ultra lightweight real-time object detection model that includes four different sizes of XS/S/M/L. 
By using structures such as TAL, ETA Head, and PAN, the accuracy of the model is improved. When deploying model inference, 
the model supports including post-processing in the network, thereby supporting direct output of prediction results.

git clone https://github.com/PaddlePaddle/PaddleDetection.git
cd PaddleDetection
pip3 install -r requirements.txt -i https://pypi.tuna.tsinghua.edu.cn/simple
python3 setup.py install
pip3 install protobuf==3.20.3
pip3 install numba==0.56.4
yum install mesa-libGl -y

Download the Coco datasets, Specify coco2017 to your Coco path in later training process. 
The Coco datesets path structure should look like(assuming coco2017 is used):

coco2017
├── annotations
│   ├── instances_train2017.json
│   ├── instances_val2017.json
│   └── ...
├── train2017
│   ├── 000000000009.jpg
│   ├── 000000000025.jpg
│   └── ...
├── val2017
│   ├── 000000000139.jpg
│   ├── 000000000285.jpg
│   └── ...
├── train2017.txt
├── val2017.txt
└── ...

export CUDA_VISIBLE_DEVICES=0
python3 tools/train.py -c configs/picodet/picodet_l_640_coco_lcnet.yml --eval

export CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7
python3 -m paddle.distributed.launch --gpus 0,1,2,3,4,5,6,7 tools/train.py -c configs/picodet/picodet_l_640_coco_lcnet.yml --eval