zitkat/potr

Pose Transformer

POTR

Pose Transformer

Standard POTR

Deformable POTR

The implementation is based on this implementation of Deformable DETR.

Building

Because the original implementation structures the MSDeformAttention module as a separate Python package with its own Python, CUDA, and C++ scripts, one needs to build this before running the training.

First create a dedicated Conda environment with Python 3.7:

conda create -n deformable_potr python=3.7 pip
conda activate deformable_potr

Then install all the dependencies using:

pip install -r requirements.txt
conda install cudatoolkit-dev -c conda-forge

And finally, build the MSDeformAttention module (and all other custom CUDA operators) using:

cd ./deformable_potr/models/ops
sh ./make.sh

You can then test the CUDA operators by running python test.py, where all tests should pass.

To get notebooks install juyptext

pip install jupytext

How to get Jupyter notebooks

To facilitate nice Git workflow we use Jupytext. See their manual for installation instruction and when creating your own notebooks.

To get Jupyter Notebook .ipynb out of a .py file using Jupyter Lab:

1. Have Jupytext installed.
2. Run Jupyter Lab.
3. In navigation sidebar right-click on a .py file, select Open With -> Notebook.
4. Save the just opened notebook. New file with .ipynb extension should appear.
5. Enjoy the new notebook, don't forget to close .py file opened as notebook ;-)