Paper: FCOS:Fully Convolutional One-Stage Object Detection(ICCV 2019)
FCOS is recognized as a nice anchor-free, one-stage detector because of its simple structure and high performance efficiency. This repo is aimed to reach a fast, pure implementation for FCOS. All of you can develop algorithms based on this repo.
you can download the pretrained model by CSDN to avoid annoying BaiduPan, click here to download. (Note that 5 C-points is needed. If you have problems with C-point, you can contact my CSDN account 木盏 to reach model for free.)
After downloading fcos_pretrained_model.pth, please put it to folder 'checkpoint', and put the images you want for test to folder 'test_images'
python detect.py
Then, you can find results at 'out_images/':
