Detectron2 - Training & Testing balloon dataset

[junxnone]

Reference

• [ ] Detectron2 Tutorial notebook colab
• [ ] Detectron2

Brief

Environment

mkvirtualenv -p python3 dtr
workon dtr
pip install torch torchvision
pip install cython pyyaml==5.1
pip install -U 'git+https://github.com/cocodataset/cocoapi.git#subdirectory=PythonAPI'
pip install detectron2 -f https://dl.fbaipublicfiles.com/detectron2/wheels/cu100/index.html

Download balloon Dataset

wget https://github.com/matterport/Mask_RCNN/releases/download/v2.1/balloon_dataset.zip
unzip balloon_dataset.zip > /dev/null

Train

• train code

Tuning

• model - COCO-InstanceSegmentation/mask_rcnn_R_50_FPN_3x.yaml
• learning rate - cfg.SOLVER.BASE_LR = 0.005
• total interations - cfg.SOLVER.MAX_ITER = 500
• learning rate reduce steps - cfg.SOLVER.STEPS = (300, 350, 400, 450)

[junxnone]

339