Object detection system trained by hybrid-supervision/weakly semi-supervision (HSOD/WSSOD):
This project is based on CenterNet.
Contact: mukai@buaa.edu.cn. Any questions or discussions are welcomed!
Pleases follow the installation instructions INSTALL.MD of CenterNet and check your pytorch version.
Pytorch=1.4, Cudatoolkit=10.1 is highly recommended.
After installation, follow the instructions in DATA.md to setup the datasets.
Run python start.py ctdet
in src/ folder, then open http://127.0.0.1:9766/
in browser.
Video demo is available at bilibili.
- Weakly-supervised training
CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7 python -m torch.distributed.launch --nproc_per_node=8 main.py ctdet --dataset=coco80 --arch dlav0camsplit_34 --exp_id camsplit_weak_train --weak --lr_step 30 --num_epochs 50
- Fully-supervised training
CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7 python -m torch.distributed.launch --nproc_per_node=8 main.py ctdet --dataset=coco35 --arch dlav0camsplit_34 --load_model ../exp/ctdet/camsplit_weak_train/model_last.pth --exp_id=camsplit_train --batch_size 64
Be aware of settingself.weak = False
in src/lib/models/networks/{arch_name}.py - Testing on coco val2017 set
python test.py ctdet --exp_id coco_test --keep_res --load_model ../exp/ctdet/camsplit_train/model_last.pth --arch dlav0camsplit_34
- Detection inference of image/folder
CUDA_VISIBLE_DEVICES=0 python demo.py ctdet --demo ./images/ --load_model ../exp/ctdet/camsplit_train/model_last.pth --arch dlav0camsplit_34 --nms
CenterNet itself is released under the MIT License (refer to the LICENSE file for details). Portions of the code are borrowed from human-pose-estimation.pytorch (image transform, resnet), CornerNet (hourglassnet, loss functions), dla (DLA network), DCNv2(deformable convolutions), tf-faster-rcnn(Pascal VOC evaluation) and kitti_eval (KITTI dataset evaluation). Please refer to the original License of these projects (See NOTICE).