This repository provides dataset classes and some utility functions for the relevant datasets, as well as a place to develop and submit your implementation of the proposed method.
- Docker and NVIDIA Container Toolkit.
- An RTX 40XX or 30XX GPU
- Our NVMe server avola mounted on /home/nfs/inf6/data/....
- Clone this repository.
- Build the project. Be aware that the transfer of user permissions at the end of the build takes about 30 minutes.
cd denoising_diffusion/Docker && ./build.sh
- Start and attach to a tmux session.
tmux
- Run a container.
./run.sh
- (Optional) For quick prototyping setup jupyter notebook in the container.
jupyter notebook password
cd /repos
jupyter notebook --no-browser --ip 0.0.0.0 --port 9999 &
(Optional) Connect to the machine hosting the container, for example using (adapt to your USER):
ssh -L 9999:localhost:9999 -J USER@login-stud.informatik.uni-bonn.de USER@robo4
In either case, the notebook is now available locally at http://localhost:9999/.
Sanity-checking the proposed method on FashionMNIST is reasonable, please see FashionMNIST_Dataset.ipynb on how to get started with that dataset. For the Binpicking scene dataset, please see BinsceneA_Dataset.ipynb.