openpifpaf_apollocar3d

Setup

Download openpifpaf, switch to dev branch and install it with:

pip3 install --editable '.[dev,train,test]

pip3 install pandas

pip3 install itermplot

(in case CUDA 9 as driver: pip install torch==1.7.0+cu92 torchvision==0.8.1+cu92 -f https://download.pytorch.org/whl/torch_stable.html)

• Download openpifpaf_animalpose, and create the following directories:
  • mkdir data
  • soft link output directory, which needs to be called outputs
  • soft link to animalpose dataset
  • create apollo-coco directory with images/train, images/val, annotations subdirectories and soft link them.

Preprocess Dataset

python -m openpifpaf_animalpose.voc_to_coco Use the argument --split_images to create a training val split copying original images in the new folders

Show poses

python -m openpifpaf_apollocar3d.utils.constants

Pretrained models

TODO

Train

TODO

Everything else

All pifpaf options and commands still hold, please check the DEV guide