yolact (2020-02-11)

Uses YOLACT for object detection (masks).

Uses PyTorch 1.2 and CUDA 10.0.

Works on 1080 Ti cards.

Version

YOLACT github repo hash (commit 322):

f54b0a5b17a7c547e92c4d7026be6542f43862e7

Timestamp:

2020-02-11

Docker

Quick start

• Log into registry using public credentials:

  docker login -u public -p public public.aml-repo.cms.waikato.ac.nz:443 
  

• Pull and run image (adjust volume mappings -v):

  docker run --gpus=all --shm-size 8G \
    -v /local/dir:/container/dir \
    -it public.aml-repo.cms.waikato.ac.nz:443/yolact/yolact:2020-02-11
  

  NB: For docker versions older than 19.03 (docker version), use --runtime=nvidia instead of --gpus=all.

• If need be, remove all containers and images from your system:

  docker stop $(docker ps -a -q) && docker rm $(docker ps -a -q) && docker system prune -a
  

Build local image

• Build image yolact from Docker file (from within yolact-2020-02-11):

  docker build -t yolact .
  

• Run image yolact in interactive mode (i.e., using bash) as container yolact_container:

  docker run --gpus=all --name yolact_container -ti -v \
    /local/dir:/container/dir \
    yolact bash
  

Pre-built images

• Build

  docker build -t yolact/yolact:2020-02-11 .
  

• Tag

  docker tag \
    yolact/yolact:2020-02-11 \
    public-push.aml-repo.cms.waikato.ac.nz:443/yolact/yolact:2020-02-11
  

• Push

  docker push public-push.aml-repo.cms.waikato.ac.nz:443/yolact/yolact:2020-02-11
  

  If error no basic auth credentials occurs, then run (enter user/password when prompted):

  docker login public-push.aml-repo.cms.waikato.ac.nz:443
  

• Pull

  If image is available in aml-repo and you just want to use it, you can pull using following command and then run.

  docker pull public.aml-repo.cms.waikato.ac.nz:443/yolact/yolact:2020-02-11
  

  If error no basic auth credentials occurs, then run (enter user/password when prompted):

  docker login public-push.aml-repo.cms.waikato.ac.nz:443
  

  Then tag by running:

  docker tag \
    public.aml-repo.cms.waikato.ac.nz:443/yolact/yolact:2020-02-11 \
    yolact/yolact:2020-02-11
  

• Run

  docker run --gpus=all --shm-size 8G -v /local/dir:/container/dir -it yolact/yolact:2020-02-11
  

  /local/dir:/container/dir maps a local disk directory into a directory inside the container. Typically, you would map the weights (pre-trained models) and the data (annotations, log, etc):

  -v /some/where/dataset01:/data -v /some/where/yolact/pretrained:/yolact/weights
  

Usage

• Instead of modifying the /yolact/data/config.py file itself, an external Python module can be loaded in via the YOLACT_CONFIG environment variable. This module must have the following two variables defined:

  • external_dataset
  • external_config

  These will get mapped by the modified config.py in the image to external_dataset and external_config.

  See external_config_example.py for an example module.

  For example, if your configuration will be available in the docker container as /data/config/model-01.py then you export the following environment variable:

  export YOLACT_CONFIG=/data/config/model-01.py
  

• Train

  yolact_train --config=external_config --log_folder=/data/log \
    --validation_epoch 100
  

• Evaluate

  yolact_eval --config=external_config --trained_model=weights/MODELNAME.pth \
    --score_threshold=0.15 --top_k=200 \
    --images=/predictions/in/:/predictions/out/
  

• Predict

  yolact_predict --config=external_config --trained_model=weights/MODELNAME.pth \
    --score_threshold=0.15 --top_k=200 \
    --output_polygons --output_minrect \
    --prediction_in /predictions/in/ --prediction_out /predictions/out/    
  

Additional configurations

• You can use YOLACT_CONFIG2 and YOLACT_CONFIG3 to supply two more configurations.

• Use --config=external_config2 and --config=external_config3 respectively when referring to them.

• Within the configuration file itself, by sure to reference the dataset correctly via 'dataset': external_dataset2, or 'dataset': external_dataset2,

Permissions

When running the docker container as regular use, you will want to set the correct user and group on the files generated by the container (aka the user:group launching the container):

docker run -u $(id -u):$(id -g) ...