Keuper Labs
← All publications

ECCV 26 · 2026

Decoupling High and Low Frequencies for Faithful Image Generation with Fine Details

Tejaswini Medi, Hsien-Yi Wang, Arianna Rampini, Margret Keuper

Generative ModelsAutoencodersFrequency AnalysisDiffusion Models

Why this publication matters

Fine details can vanish when a generator stores an image in a compressed internal form. DeBaT separates broad structure from high-frequency detail so that both receive dedicated attention. It improves the preservation of texture and small features while keeping the overall image coherent.

Abstract

Latent generative models compress images into learned embeddings prior to synthesis, and the generation quality critically depends on how faithfully these embeddings preserve visual detail. We observe that while such embeddings are effective at reconstructing low frequency structure, they struggle to recover sharp high frequency details that are essential for perceptual realism. Conventional reconstruction objectives implicitly prioritize coarse structural information over high frequency content, which can lead to overly smoothed outputs and degraded visual quality in textured regions. Motivated by this observation, we propose DeBaT, a Decoupled frequency Band Tokenizer that explicitly separates the learning of low and high frequency band embeddings. This decoupling enables accurate reconstruction of fine details while preserving global coherence. Integrated into a latent diffusion based generative model, DeBaT allows for sharper and more realistic samples than previous latent tokenizers, confirming that the explicit decoupling of high and low frequency bands eases the preservation of visual details in learned embedding spaces.

Abstract source ↗

Figures

Fine-detail reconstruction crops comparing DeBaT with a baseline tokenizer.
Figure 1. Reconstruction quality for two examples. We compare the reconstruction of VA-VAE to our approach DeBaT. Highlighted regions emphasize textures and edges. DeBaT better preserves high-frequency details and sharp structures, producing reconstructions visually closer to the input. Arrows indicate reconstruction errors. View in source ↗
Comparison of wavelet based latent VAE tokenizer designs.
Figure 2. Comparison of wavelet based latent VAE tokenizer designs. Previous works for image (VA-VAE) and video (WF-VAE) generation leverage spatial compression properties of multiple wavelet stages, while low and high frequency bands are coupled in a joint latent code. We propose DeBaT, which preserves fine details by learning explicit latent codes for low and for high frequency bands through decoupled VAEs. View in source ↗

Cite this paper

Download .bib 
@inproceedings{medi2026decouplinghighand93,
  title = {{Decoupling High and Low Frequencies for Faithful Image Generation with Fine Details}},
  author = {Tejaswini Medi and Hsien-Yi Wang and Arianna Rampini and Margret Keuper},
  booktitle = {European Conference on Computer Vision},
  year = {2026},
  url = {https://arxiv.org/abs/2509.05441v4}
}

Figures and abstract are reproduced from the linked research sources. Credit remains with the authors and publishers.