detectron docker image from Hang

gistfile1.md

Getting docker running on CentOS server

   sudo systemctl start docker
   sudo systemctl status docker
   sudo systemctl enable docker
   sudo usermod -aG docker $(whoami)

Check if it works: sudo docker run hello-world

Getting Hang's image of Detectron

• sudo docker run -v /srv/data1/arunirc/:/detectron/data1 -e NVIDIA_VISIBLE_DEVICES=0 --runtime=nvidia --rm -it suhangpro/detectron /bin/bash
• sudo docker run -v /srv/data1/arunirc/:/detectron/data1 --runtime=nvidia --rm -it suhangpro/detectron /bin/bash

Using my own docker container for Detectron

• sudo docker run -v /srv/data1/arunirc/:/detectron/data1 --runtime=nvidia --rm -it arunirc/detectron-perturb /bin/bash

Commit changes to docker image

• Get the docker ID of a running docker container: docker ps

• sudo docker commit CONTAINER_ID arunirc/detectron-perturb

MS-COCO eval + training using Docker image of Detectron

• Inside the docker container, create symlink to downloaded COCO folder from /detectron/lib/datasets/data.
• Check the COCO paths inside /detectron/lib/datasets/dataset_catalog.py and confirm they match your COCO folder inside the docker.
• Run evaluation on COCO using Detectron example:

N=8

python2 tools/test_net.py \
    --cfg configs/12_2017_baselines/e2e_mask_rcnn_R-101-FPN_2x.yaml \
    --multi-gpu-testing \
    TEST.WEIGHTS https://s3-us-west-2.amazonaws.com/detectron/35861858/12_2017_baselines/e2e_mask_rcnn_R-101-FPN_2x.yaml.02_32_51.SgT4y1cO/output/train/coco_2014_train:coco_2014_valminusminival/generalized_rcnn/model_final.pkl \
    NUM_GPUS $N

• Run evaluation on specified Retinanet_R-50-FPN model

N=8
CFG_PATH=configs/12_2017_baselines/retinanet_R-50-FPN_2x.yaml
WT_PATH=/detectron/data1/Research/RetinaNet-COCO/data/models/retinanet_R-50-FPN_2x.pkl

python2 tools/test_net.py \
    --cfg ${CFG_PATH} \
    --multi-gpu-testing \
    TEST.WEIGHTS ${WT_PATH} \
    NUM_GPUS $N