Summary of Scalable Image Tokenization with Index Backpropagation Quantization, by Fengyuan Shi et al.
Scalable Image Tokenization with Index Backpropagation Quantization
by Fengyuan Shi, Zhuoyan Luo, Yixiao Ge, Yujiu Yang, Ying Shan, Limin Wang
First submitted to arxiv on: 3 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 proposed Index Backpropagation Quantization (IBQ) method addresses the scalability issue in existing vector quantization (VQ) methods by optimizing all codebook embeddings and visual encoders jointly. By applying a straight-through estimator on one-hot categorical distributions, IBQ enables scalable training of visual tokenizers with large-scale codebooks. The approach achieves competitive results on reconstruction and autoregressive visual generation tasks on the ImageNet benchmark. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary IBQ is a new way to optimize vector quantization for image recognition. It helps machines learn to compress images more efficiently and accurately. This method overcomes previous limitations by allowing all parts of the codebook to be updated during training, making it much faster and more reliable. The results show that IBQ can do just as well as other methods on large datasets. |
Keywords
» Artificial intelligence » Autoregressive » Backpropagation » One hot » Quantization
