Saved in:
Bibliographic Details
Main Author: Jin, Haopeng
Format: Preprint
Published: 2026
Subjects:
Online Access:https://arxiv.org/abs/2604.22808
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866918466605285376
author Jin, Haopeng
author_facet Jin, Haopeng
contents Long-sequence video diffusion transformers hit a quadratic self-attention cost that dominates runtime and memory for very long token sequences. Most efficient attention methods use one approximation everywhere, yet video features are spectrally structured: low frequencies carry global layout and coarse motion; high frequencies carry texture and fine detail. We present FreqFormer, a frequency-aware heterogeneous attention framework. Token features are split into spectral bands with different operators: dense global attention on compressed low-frequency content, structured block-sparse attention on mid frequencies, and sliding-window local attention on high frequencies. A lightweight spectral routing network allocates heads across bands using layer statistics and the diffusion timestep, shifting compute toward global structure early in denoising and detail later. Cross-band summary tokens provide cheap residual exchange. FreqFormer is paired with a fused GPU execution plan that co-schedules dense, sparse, and local branches to cut kernel launches and memory traffic. We give a consistent complexity model, an orthonormal-decomposition view of approximation, and simulation-based systems numbers (throughput, arithmetic intensity, memory traffic, duration scaling). In simulations from 64K to 1M tokens, FreqFormer substantially reduces estimated attention FLOPs and KV-related memory traffic versus dense attention while keeping a hardware-friendly pattern, supporting spectrally structured heterogeneous attention as a practical direction for long-video diffusion transformers.
format Preprint
id arxiv_https___arxiv_org_abs_2604_22808
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle FreqFormer: Hierarchical Frequency-Domain Attention with Adaptive Spectral Routing for Long-Sequence Video Diffusion Transformers
Jin, Haopeng
Computer Vision and Pattern Recognition
Artificial Intelligence
Image and Video Processing
68T05, 65F10
I.2.6; G.1.0
Long-sequence video diffusion transformers hit a quadratic self-attention cost that dominates runtime and memory for very long token sequences. Most efficient attention methods use one approximation everywhere, yet video features are spectrally structured: low frequencies carry global layout and coarse motion; high frequencies carry texture and fine detail. We present FreqFormer, a frequency-aware heterogeneous attention framework. Token features are split into spectral bands with different operators: dense global attention on compressed low-frequency content, structured block-sparse attention on mid frequencies, and sliding-window local attention on high frequencies. A lightweight spectral routing network allocates heads across bands using layer statistics and the diffusion timestep, shifting compute toward global structure early in denoising and detail later. Cross-band summary tokens provide cheap residual exchange. FreqFormer is paired with a fused GPU execution plan that co-schedules dense, sparse, and local branches to cut kernel launches and memory traffic. We give a consistent complexity model, an orthonormal-decomposition view of approximation, and simulation-based systems numbers (throughput, arithmetic intensity, memory traffic, duration scaling). In simulations from 64K to 1M tokens, FreqFormer substantially reduces estimated attention FLOPs and KV-related memory traffic versus dense attention while keeping a hardware-friendly pattern, supporting spectrally structured heterogeneous attention as a practical direction for long-video diffusion transformers.
title FreqFormer: Hierarchical Frequency-Domain Attention with Adaptive Spectral Routing for Long-Sequence Video Diffusion Transformers
topic Computer Vision and Pattern Recognition
Artificial Intelligence
Image and Video Processing
68T05, 65F10
I.2.6; G.1.0
url https://arxiv.org/abs/2604.22808