Summary of Lmlt: Low-to-high Multi-level Vision Transformer For Image Super-resolution, by Jeongsoo Kim et al.
LMLT: Low-to-high Multi-Level Vision Transformer for Image Super-Resolution
by Jeongsoo Kim, Jongho Nang, Junsuk Choe
First submitted to arxiv on: 5 Sep 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 recent surge in Vision Transformer (ViT)-based methods for Image Super-Resolution has led to impressive performance gains. However, these models suffer from significant complexity, resulting in high inference times and memory usage. Additionally, ViT models using Window Self-Attention (WSA) face challenges in processing regions outside their windows. To address these issues, the proposed Low-to-high Multi-Level Transformer (LMLT) employs attention with varying feature sizes for each head. LMLT divides image features along the channel dimension, gradually reduces spatial size for lower heads, and applies self-attention to each head. This approach effectively captures both local and global information. By integrating the results from lower heads into higher heads, LMLT overcomes the window boundary issues in self-attention. Extensive experiments demonstrate that our model significantly reduces inference time and GPU memory usage while maintaining or even surpassing the performance of state-of-the-art ViT-based Image Super-Resolution methods. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary The paper proposes a new method for improving image super-resolution, which is important because it can help us make images look better. The old way of doing this uses something called Vision Transformers (ViTs), but these are very complex and take up a lot of computer memory. They also have trouble looking at parts of the picture that are outside their “window”. To fix these problems, the researchers came up with a new idea called Low-to-high Multi-Level Transformer (LMLT). This works by breaking down the image into smaller parts, looking at each part in different ways, and then putting all those parts back together again. It’s like taking a puzzle apart, looking at each piece in a different way, and then putting it all back together to make a complete picture. |
Keywords
» Artificial intelligence » Attention » Inference » Self attention » Super resolution » Transformer » Vision transformer » Vit
