Eztorch

Published:

About

Eztorch is a library to make training, validation, and testing in PyTorch easy to perform image and video self-supervised representation learning and evaluate those representations on downstream tasks.

It encapsulates contributions from my Ph.D. research which reduced training time by up to 50% in comparison with other researchers’ code at the time.

It led to the following academic contributions:

  • Similarity Contrastive Estimation for Self-Supervised Soft Contrastive Learning (WACV 2023)
  • Similarity Contrastive Estimation for Image and Video Soft Contrastive Self-Supervised Learning (MVAP 2023)
  • COMEDIAN: Self-Supervised Learning and Knowledge Distillation for Action Spotting using Transformers (WACV Workshops 2024)

To ease the use of the code, documentation has been built.

Github

Stack

PyTorch, Github Actions