jaswar / cvdl-project

Computer Vision by Deep Learning Project - Group 39

Authors: Maurits Kienhuis, Niels van der Voort, Jan Warchocki

Prerequisites

To install only the packages required for generating datasets, run the following (preferably in a virtual environment):

pip install -r requirements_datasets.txt

If you wish to install the packages required for running the models as well, run the following:

pip install -r requirements_all.txt

Generating datasets

The datasets can be generated by running the following:

python generators.py --experiment=<experiment> --seq_len=<seq_len>

Where experiment can be one of pendulum, pendulum_scale, or pendulum_intensity. The parameter seq_len is the number of frames that should be generated for each sample. For example, Physics-as-Inverse-Graphics uses seq_len=12 for training and seq_len=30 for testing.

Running the models

The instructions for running the model can be found in the README.md file in the corresponding directories.

Please note that the Physics-as-Inverse-Graphics model also requires the path to where the datasets are located. This can be specified using the flag --data_dir=<path>. The default path is ../../data/datasets/.

Credits

Physics-as-Inverse-Graphics:

• Original paper: https://arxiv.org/abs/1905.11169
• Original code: https://github.com/seuqaj114/paig
• Tensorflow 2 implementation: https://github.com/Alejandro-neuro/paig_reproducibility