Summary of Learning to Merge Tokens Via Decoupled Embedding For Efficient Vision Transformers, by Dong Hoon Lee et al.
Learning to Merge Tokens via Decoupled Embedding for Efficient Vision Transformers
by Dong Hoon Lee, Seunghoon Hong
First submitted to arxiv on: 13 Dec 2024
Categories
- Main: Computer Vision and Pattern Recognition (cs.CV)
- Secondary: Artificial Intelligence (cs.AI)
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 paper proposes Decoupled Token Embedding for Merging (DTEM), a method that enhances token merging in Vision Transformers (ViTs) by learning dedicated features for token merging. The approach introduces a lightweight embedding module decoupled from the ViT forward pass, which is learned via a continuously relaxed token merging process. This allows for differentiable learning of the decoupled embeddings, making it possible to integrate DTEM into existing ViT backbones and train it modularly or end-to-end. The authors demonstrate the applicability of DTEM on various tasks, including classification, captioning, and segmentation, with consistent improvement in token merging. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary The paper introduces a new method for improving Vision Transformers (ViTs) called Decoupled Token Embedding for Merging (DTEM). This helps to make ViTs better at combining similar tokens into fewer ones. The authors created a special module that can be added to existing ViT models and trained separately or as part of the entire model. They tested DTEM on different tasks like image classification, captioning, and segmentation and found it improved performance. |
Keywords
» Artificial intelligence » Classification » Embedding » Image classification » Token » Vit
