FCOS pytorch implementation based on detectron2
This is repository of my re-implementation code of FCOS: Fully Convolutional One-Stage Object Detection - ICCV, 2019, pp. 9627-9636
Requirement
Ubuntu 16.04
Pytorch >= 1.3
torchvision
pycocotools $ pip install cython; pip install -U'git+https://github.com/cocodataset/cocoapi.git#subdirectory=PythonAPI’
Build detectron2 from source
# install it from a local clone:
$ git clone https://github.com/facebookresearch/detectron2.git
$ cd detectron2 && python -m pip install -e .
Training with coco dataset(4 GPUs)
python train_net.py --num-gpus 4 --config-file configs/R_50_1x.yaml
Training with coco dataset(1 GPUs)
python train_net.py --config-file configs/R_50_1x.yaml \
--num-gpus 1 SOLVER.IMS_PER_BATCH 4 SOLVER.BASE_LR 0.0025
Run demo with trained model
python train_net.py --config-file configs/R_50_1x.yaml \
--eval-only MODEL.WEIGHTS /path/to/checkpoint_file
Code reference
https://github.com/aim-uofa/AdelaiDet
https://github.com/facebookresearch/detectron2