Summary of Nested-tnt: Hierarchical Vision Transformers with Multi-scale Feature Processing, by Yuang Liu et al.
Nested-TNT: Hierarchical Vision Transformers with Multi-Scale Feature Processing
by Yuang Liu, Zhiheng Qiu, Xiaokai Qin
First submitted to arxiv on: 20 Apr 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 Transformer architecture has been successfully applied to computer vision tasks, outperforming traditional convolutional neural networks (CNNs) and achieving new state-of-the-art results. The Vision Transformers (ViT) approach divides images into local patches, or “visual sentences,” but this alone is insufficient to capture the vast complexity of image features. To address this limitation, the TNT model is proposed, which further subdivides the image into smaller patches, or “visual words,” leading to more accurate results. The core mechanism of Transformer is Multi-Head Attention, which traditional attention mechanisms ignore, leading to redundancy and underutilization. To mitigate these issues, a nested algorithm is introduced, and the Nested-TNT model is applied to image classification tasks. Experimental results demonstrate that the proposed model achieves better performance than ViT and TNT on datasets CIFAR10 and FLOWERS102, exceeding 2.25%, 1.1% and 2.78%, 0.25% respectively. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Transformer helps computer vision by doing a great job in natural language processing. It’s been used to look at pictures too! The problem is that looking at small parts of the picture isn’t enough, we need to consider how those parts relate to each other. To solve this, a new way of looking at pictures is proposed called TNT. This helps us get even better results. The core idea is called Multi-Head Attention and it’s what makes Transformer so powerful. We use this to look at lots of tiny pieces of the picture and figure out how they fit together. It seems to work really well! |
Keywords
* Artificial intelligence * Attention * Image classification * Multi head attention * Natural language processing * Transformer * Vit
