Image Generation with Channel-wise Quantization
Keywords: generative models, image generation, visual tokenization
Abstract: We present a novel image generation model with channel-wise quantization. Our method quantizes image feature along channel into discrete codes. Then based on the learned codes, our approach adopts masked-prediction paradigm for image generation. Compared with widely used spatial tokenizers, our channel-wise tokenizer has an efficient modeling for image structure and strong representational capacity. Besides, the codebook usage of our tokenizer can reach 100\% under different codebook size. Using the channel-wise tokenizer, our generation framework achieves competitive performances on various benchmarks of image generation. In particular, on ImageNet 256x256 benchmark, our method significantly improve baseline by improving Frechet inception distance (FID) to 1.87. Furthermore, we also validate the effectiveness of our proposed method on text-to-image generation.
Primary Area: generative models
Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.
Reciprocal Reviewing: I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6.
Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.
No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.
Submission Number: 1432
Loading