Image Tokenizers Should Be Trained Between Images
Latent diffusion models generate images as codes of an image tokenizer, whose decoder turns them into the pixel image.
ProofPaper ↗
Key points
- Recent tokenizers improve how well these codes reconstruct an image and how easily a generator learns them.
- To improve generated images, decoders should therefore be trained on codes in the space between training samples.
- We propose InterpTok, which fine-tunes MAETok to decode mixtures of image codes into natural images.
- Beyond interpolation, InterpTok improves the generation FID of a matched diffusion transformer from 11.61 to 8.48 without guidance, and even reaches 8.11 at an eighth of the sampling steps, where the generator on MAETok's codes worsens.
Sources (2)
- [1]Image Tokenizers Should Be Trained Between ImagesarXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 12, 01:23 PM
Latent diffusion models generate images as codes of an image tokenizer, whose decoder turns them into the pixel image.
Recent tokenizers improve how well these codes reconstruct an image and how easily a generator learns them.
- [2]Image Tokenizers Should Be Trained Between ImagesHugging Face Daily Papers · Oct 12, 01:23 PM · same content
Extractive summary: sentences quoted from the sources.
Before this
- Oct 11, 2026TabMosaic: Unified In-Context Diffusion for Tabular Generative Tasks
- Oct 11, 2026AesCode: Aesthetic Code Generation with Decoupled Cross-Modal Rewards
- Oct 11, 2026Imperfect Verifiers, Exact Targets: Inference-Time Scaling for Continuous Generation
- Oct 11, 2026Data-Centric Recursive Self-Improvement for Navigation via Policy-Synchronized Language-Action Dual Bootstrapping
- Oct 10, 2026Compositional Concepts Erasing: Disentangling Object-Style via Domain Manifold Unlearning
- Oct 10, 2026LightVAE: Towards Compact and Efficient Video Autoencoders