TileShare speeds up vision transformer decoding by reducing overlap work
TileShare: Training-Free Acceleration of Overlapping Tiled ViT Decoding
Computer Vision and Pattern RecognitionMachine Learning
Summary
Vision Transformer (ViT) decoders process images or videos by breaking them into overlapping tiles, but this causes repeated work when decoding overlapping parts. The authors found a way to share some computations across these overlapping tiles without retraining the model, cutting duplicated effort. This speeds up decoding by up to 1.96 times while keeping image quality nearly identical. They tested this on video and image models, showing it works broadly and even improves some quality measures on unseen data.
What this means in practice
- •For video processing teams: Speed up video decoder pipelines by sharing computations in overlapping tile regions without retraining decoders.
- •For image processing engineers: Accelerate high-resolution image autoencoders decoding with no loss in quality by reducing redundant tile computations.
Authors
Xiangzhong Ye, Zhongshen Li, Shen You, Ziyang Meng, Liang Xu, Junkai Ji, Qiuzhen Lin, Xiangtao Li, Ka-Chun Wong
Abstract
Fixed-window Vision Transformer (ViT) decoders run at deployment resolution as overlapping tiles: every token in an overlap band is processed by several full decoder instances. In the 2.4B-parameter MiniMax-H3 video decoder, about half of all per-token computation is duplicated. Removing the tiles fails because attention, global tokens and coordinates are bound to them. TileShare keeps each released tile's attention and computes every other operation once per token. The released computation changes at one kind of step only: an overlap token's per-tile results are blended into one shared state with the released weights. Interventions show that this blend costs quality where attention reads the shared states, mainly in the first few layers, which TileShare therefore keeps per tile. Without training, keeping six layers per tile decodes 1.38-1.51x faster than batched and compiled tiling on two GPU types (up to 1.96x with FP8) and is equivalent to the released decoder in PSNR, LPIPS and DISTS on DAVIS videos. On unseen UVG videos with the setting frozen beforehand, it matches PSNR and DISTS and improves LPIPS. TileShare carries over to two further decoders and is again confirmed on unseen data. It is equivalent in PSNR and DISTS on MAGI-1. On the ViTok-v2 image autoencoder at 2K-4K, it matches overlapping tiling in PSNR, improves LPIPS and DISTS and decodes 1.5-1.8x faster. How many layers stay per tile is measured once per model (for decoders, without labels); we report it for 14 decoders and encoders. Optionally, two GPU-hours of self-distillation let MiniMax-H3 share every layer without a loss in PSNR, and a coordinate field trained while the network stays frozen brings fully shared CLIP-based open-vocabulary segmentation to parity with sliding windows on seven benchmarks.