Summary of Temporally Consistent Unbalanced Optimal Transport For Unsupervised Action Segmentation, by Ming Xu et al.
Temporally Consistent Unbalanced Optimal Transport for Unsupervised Action Segmentation
by Ming Xu, Stephen Gould
First submitted to arxiv on: 1 Apr 2024
Categories
- Main: Computer Vision and Pattern Recognition (cs.CV)
- Secondary: Machine Learning (cs.LG); Image and Video Processing (eess.IV)
GrooveSquid.com Paper Summaries
GrooveSquid.com’s goal is to make artificial intelligence research accessible by summarizing AI papers in simpler terms. Each summary below covers the same AI paper, written at different levels of difficulty. The medium difficulty and low difficulty versions are original summaries written by GrooveSquid.com, while the high difficulty version is the paper’s original abstract. Feel free to learn from the version that suits you best!
| Summary difficulty | Written by | Summary |
|---|---|---|
| High | Paper authors | High Difficulty Summary Read the original abstract here |
| Medium | GrooveSquid.com (original content) | Medium Difficulty Summary The proposed method solves an optimal transport problem to segment long, untrimmed videos into actions. It encodes a temporal consistency prior into a Gromov-Wasserstein problem and decodes a temporally consistent segmentation from a noisy affinity/matching cost matrix between video frames and action classes. Unlike previous approaches, the method does not require knowing the action order for a video to attain temporal consistency. The resulting Gromov-Wasserstein problem can be efficiently solved on GPUs using projected mirror descent. The method is demonstrated in an unsupervised learning setting, generating pseudo-labels for self-training. It achieves state-of-the-art results on the Breakfast, 50-Salads, YouTube Instructions, and Desktop Assembly datasets. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary The paper proposes a new way to segment actions from long videos without knowing the order of the actions. This is done by solving an optimal transport problem that takes into account how similar each video frame is to different actions. The method is good at separating different actions in videos even when there are many actions and they happen close together. It’s used for unsupervised learning, where it generates labels for self-training. This approach achieves the best results on several popular datasets. |
Keywords
» Artificial intelligence » Self training » Unsupervised
