PP-PicoDet: A Better Real-Time Object Detector on Mobile Devices

Introduction

latex @article{picodet, title={{PP-PicoDet}: A Better Real-Time Object Detector on Mobile Devices}, author={Guanghua Yu, Qinyao Chang, Wenyu Lv, Chang Xu, Cheng Cui, Wei Ji, Qingqing Dang, Kaipeng Deng, Guanzhong Wang, Yuning Du, Baohua Lai, Qiwen Liu, Xiaoguang Hu, Dianhai Yu, Yanjun Ma}, journal={arXiv preprint arXiv:2111.00902}, year={2021} }

Backbone Pretrained Weights

• ESNet_0.75
• ESNet_1.0x
• ESNet_1.25x

Results and Models

| Bakcbone | size|box AP(ppdet) | Config|Download |:---------:|:-------:|:-------:|:-------:|:-------:| |picodet-s|320|26.9(27.1)| config|model | log| |picodet-s|416|30.6(30.6)| config|model | log| |picodet-m|416|34.2(34.3)| config|model | log| |picodet-l|640|40.4(40.9)| config|model | log|

Usage

Install MMdetection

Our implementation is based on mmdetection. Install mmdetection according to INSTALL

Note: Make sure your mmcv-full version is consistency with mmdet version(we use mmcv==1.4.0)

Train

1. Download pretrained backbone using the link above

2. training

bash tools/dist_train.sh ./configs/picodet/picodet_s_320_coco.py 4

Test

``` bash tools/disttest.sh $CONFIGPATH $TRAINEDMODELPATH $GPU_NUMS --eval bbox

eg. use picodet-s 320 pretrianed model bash tools/disttest.sh ./configs/picodet/picodets320coco.py $MODELDIR/picodets_320.26.9.pth 8 --eval bbox

Evaluating bbox... Loading and preparing results... DONE (t=1.76s) creating index... index created! Running per image evaluation... Evaluate annotation type bbox DONE (t=43.50s). Accumulating evaluation results... DONE (t=14.63s).

Average Precision (AP) @[ IoU=0.50:0.95 | area= all | maxDets=100 ] = 0.269 Average Precision (AP) @[ IoU=0.50 | area= all | maxDets=1000 ] = 0.408 Average Precision (AP) @[ IoU=0.75 | area= all | maxDets=1000 ] = 0.279 Average Precision (AP) @[ IoU=0.50:0.95 | area= small | maxDets=1000 ] = 0.076 Average Precision (AP) @[ IoU=0.50:0.95 | area=medium | maxDets=1000 ] = 0.269 Average Precision (AP) @[ IoU=0.50:0.95 | area= large | maxDets=1000 ] = 0.462 Average Recall (AR) @[ IoU=0.50:0.95 | area= all | maxDets=100 ] = 0.421 Average Recall (AR) @[ IoU=0.50:0.95 | area= all | maxDets=300 ] = 0.421 Average Recall (AR) @[ IoU=0.50:0.95 | area= all | maxDets=1000 ] = 0.421 Average Recall (AR) @[ IoU=0.50:0.95 | area= small | maxDets=1000 ] = 0.138 Average Recall (AR) @[ IoU=0.50:0.95 | area=medium | maxDets=1000 ] = 0.470 Average Recall (AR) @[ IoU=0.50:0.95 | area= large | maxDets=1000 ] = 0.684 ```

Deploy

TODO： - [ ] mnn deploy