Detectron - Detectron2

[junxnone]

Reference

• [ ] detectron2 - FAIR
• [ ] Docs
• [ ] TACO using Detectron2.ipynb

Brief

Name	implementations
Detectron	Caffe2
maskrcnn-benchmark	PyTorch 1.0
Detectron2	PyTorch ≥ 1.4

Environment

Docker 使用

git clone https://github.com/facebookresearch/detectron2.git
cd  detectron2/docker
docker-compose run detectron2

Host Virtualenv

mkvirtualenv -p python3 pytorch
pip3 install torch==1.3.1+cu92 torchvision==0.4.2+cu92 -f https://download.pytorch.org/whl/torch_stable.html
pip3 install cython
pip3 install -U 'git+https://github.com/facebookresearch/fvcore.git' 'git+https://github.com/cocodataset/cocoapi.git#subdirectory=PythonAPI'
pip3 install -e detectron2/

N.B. GCC/G++ 要用 4.9

Config

• add new config

  from detectron2.config import get_cfg
  cfg = get_cfg() 
  cfg.xxx = yyy     

  cfg is CfgNode Instance

• load from config yaml file

  cfg.merge_from_file("my_cfg.yaml")   

• load from list

  cfg.merge_from_list(["MODEL.WEIGHTS", "weights.pth"])  

• dump configs

  print(cfg.dump())  

  all configs name

Data

• register dataset

  
  def get_dicts():
  ...
  return list[dict] in the following format

from detectron2.data import DatasetCatalog DatasetCatalog.register("my_dataset", get_dicts)

- register metadata

[junxnone]

291 #465 #213

TODO

• [x] Configs
• [ ] DataLoader
• [ ] Save checkpoint / model
• [ ] Evaluation
• [ ] Tuning model
• [ ] Test different model