mlsanityData to 7 Oct 2026

SpectralDiT: Timestep-Conditioned Spectral Residual Correction for Flow-Matching DiTs

arXiv 2606.18765 ↗First posted 17 Jun 2026cs.CVcited by 0

On the map: Diffusion models → Flow matching, time series and science

Abstract (arXiv)

We propose SpectralDiT, a lightweight modification to flow-matching Diffusion Transformers that adds timestep-conditioned spectral correction to the MLP residual branch. The module decomposes each residual update into low- and high-frequency components on the patch-token grid, then learns a zero-initialized additive gate so the model initially matches the baseline DiT. On CIFAR-10 pixel-space generation, SpectralDiT improves FID from 20.78 to 19.71 at patch size 1 and reduces the radial Fourier spectrum gap. Furthermore, we scale our method to latent diffusion on ImageNet-100. With 0.6% additional theoretical FLOPs and 1.36% additional parameters, SpectralDiT improves latent flow-matching, achieving an 8.7% relative FID reduction under classifier-free guidance (CFG 2.0). All reported results are averaged over five seeds. Ablations and gate visualizations on CIFAR-10 reveal stable block-specific spectral correction patterns.

Key facts. No citing papers in the map yet (data through 7 Oct 2026). Placed in Flow matching, time series and science, one of 14 directions in Diffusion models. Builds on 16 of its 23 references that are in the map. First posted on arXiv 17 Jun 2026.

Builds on

16 of 23 references.

Closest ancestors · 1

Same direction, newest first

  1. 2022Flow Matching for Generative Modeling6 Oct

Other references · 15

  1. 2020An Image is Worth 16x16 Words: Transformers for Image Recognition at Scalecited by 29,915
  2. 2020Denoising Diffusion Probabilistic Modelscited by 20,851
  3. 2021High-Resolution Image Synthesis with Latent Diffusion Modelscited by 18,694
  4. 2021Diffusion Models Beat GANs on Image Synthesiscited by 8,185
  5. 2022Scalable Diffusion Models with Transformerscited by 5,931
  6. 2022Classifier-Free Diffusion Guidancecited by 5,717
  7. 2020Fourier Features Let Networks Learn High Frequency Functions in Low Dimensional Domainscited by 2,170
  8. 2018On the Spectral Bias of Neural Networkscited by 1,321
  9. 2019Improved Precision and Recall Metric for Assessing Generative Modelscited by 925
  10. 2024SiT: Exploring Flow and Diffusion-based Generative Models with Scalable Interpolant Transformerscited by 707
  11. 2019Fourier Spectrum Discrepancies in Deep Network Generated Imagescited by 138
  12. 2025DDT: Decoupled Diffusion Transformercited by 100
3 more
  1. 2025DeCo: Frequency-Decoupled Pixel Diffusion for End-to-End Image Generationcited by 55
  2. 2025A Fourier Space Perspective on Diffusion Modelscited by 48
  3. 2025Guidance in the Frequency Domain Enables High-Fidelity Sampling at Low CFG Scalescited by 13

Built on by

Not cited yet.

Similar, not cited

Close in wording, no citation either way

  1. 2026MPDiT: Multi-Patch Global-to-Local Transformer Architecture For Efficient Flow Matching and Diffusion Model
  2. 2026Learning When to Denoise: Optimizing Asynchronous Schedules for Latent Diffusion
  3. 2026Principal-timestep Restricted Init via Sparse Matrix-decomposition in Flow-matching
  4. 2025Efficiency vs. Fidelity: A Comparative Analysis of Diffusion Probabilistic Models and Flow Matching on Low-Resource Hardware
  5. 2026Temporal Pair Consistency for Variance-Reduced Flow Matching
  6. 2026Correcting Guided Diffusion Trajectories with Spectral Alignment
  7. 2026Balancing Frequencies and Pixels in Flow Matching
  8. 2025Diffusion Bridge or Flow Matching? A Unifying Framework and Comparative Analysis