Summary of Image and Video Tokenization with Binary Spherical Quantization, by Yue Zhao et al.
Image and Video Tokenization with Binary Spherical Quantization
by Yue Zhao, Yuanjun Xiong, Philipp Krähenbühl
First submitted to arxiv on: 11 Jun 2024
Categories
- Main: Computer Vision and Pattern Recognition (cs.CV)
- Secondary: Information Theory (cs.IT); 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 A novel transformer-based tokenizer is proposed, which combines Binary Spherical Quantization (BSQ) for efficient and scalable visual data compression. The tokenizer uses a simple block-wise causal masking technique to support variable-length videos as input. The resulting BSQ-ViT achieves state-of-the-art visual reconstruction quality on image and video benchmarks with 2.4 times the throughput of prior methods. Additionally, it enables masked language models to achieve competitive image synthesis quality comparable to GAN- and diffusion-based methods. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper develops a new way to compress and reconstruct images and videos using a transformer-based tokenizer. It uses a special technique called Binary Spherical Quantization (BSQ) to make the process more efficient and scalable. The method can compress visual data by up to 100 times without losing much quality. The authors also show that their approach works well with masked language models, which are used for tasks like generating new images. |
Keywords
» Artificial intelligence » Diffusion » Gan » Image synthesis » Quantization » Tokenizer » Transformer » Vit
