Visual Self-Supervised Imitation-Learning (VSSIL)

SETUP:

• Make sure pip is installed
• Install external packages using pip install -r requirements.txt
• Install local packages with pip install -e .
• Alternatively create conda env with environment.yml

RUN:

• Example for loading the (processed) MIME data: python tests/video_dataset_test.py
  • Also includes an example for data-augmentation
• Example for the training of the Deep Spatial Auto-Encoder: python scripts/spatial_ae_training.py
• View results and training process using tensorboard --logdir results/
• The baseline from "Unsupervised Learning of Object Structure and Dynamics" can be run with different scripts, e.g. python scripts/ulosd_acrobot_training on the acrobot example data. This script takes two input arguents:
  • -c : The path to the configuration .yaml file
  • -d : The path to the video frame dataset, as specified in https://github.com/RaivoKoot/Video-Dataset-Loading-Pytorch