Summary of Language-guided Image Tokenization For Generation, by Kaiwen Zha et al.
Language-Guided Image Tokenization for Generation
by Kaiwen Zha, Lijun Yu, Alireza Fathi, David A. Ross, Cordelia Schmid, Dina Katabi, Xiuye Gu
First submitted to arxiv on: 8 Dec 2024
Categories
- Main: Computer Vision and Pattern Recognition (cs.CV)
- Secondary: Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
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 a novel image tokenization method called Text-Conditioned Image Tokenization (TexTok) that leverages language to improve compression rates and reconstruction quality. TexTok conditions the tokenization process on descriptive text captions, allowing it to focus on encoding fine-grained visual details into latent tokens. Compared to conventional tokenizers, TexTok achieves average improvements in reconstruction FID of 29.2% and 48.1% on ImageNet-256 and -512 benchmarks respectively. These improvements consistently translate to average gains in generation FID of 16.3% and 34.3%. By replacing the tokenizer in Diffusion Transformer (DiT) with TexTok, the system can achieve a 93.5x inference speedup while outperforming the original DiT using only 32 tokens on ImageNet-512. TexTok also achieves state-of-the-art FID scores of 1.46 and 1.62 on ImageNet-256 and -512 respectively. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper is about a new way to make images smaller without losing quality. It’s like taking a big photo album and turning it into a tiny box that still shows the same pictures. The method uses words to help it figure out what’s important in an image, so it can focus on keeping those details. This makes the process faster and more efficient. With this new method, images look better and take up less space. It’s like having a super-powerful photo editor that can make high-quality images appear instantly. |
Keywords
» Artificial intelligence » Diffusion » Inference » Tokenization » Tokenizer » Transformer
