OTT-Vid: Optimal Transport Temporal Token Compression for Video Large Language Models

TL;DR

OTT-Vid introduces a transport-based token compression framework for Video-LLMs that adaptively preserves semantic tokens and reduces inference costs while maintaining high performance.

Contribution

It proposes a novel optimal transport-based method for semantic-aware, adaptive temporal token compression in video models, outperforming existing methods.

Findings

Preserves 95.8% of VQA performance with only 10% tokens.

Retains 73.9% of VTG performance with 10% tokens.

Outperforms state-of-the-art training-free compression methods.

Abstract

As Video Large Language Models (Video-LLMs) scale to longer and more complex videos, their inference cost grows rapidly due to the large volume of visual tokens accumulated across frames. Training-free token compression has emerged as a practical solution to this bottleneck. However, existing temporal compression methods rely primarily on cross-frame token similarity or segmentation heuristics, overlooking each token's semantic role within its frame and failing to adapt compression strength to the compressibility of each frame pair. In this work, we propose OTT-Vid, a transport-derived allocation framework for temporal token compression. Our approach consists of two stages: spatial pruning identifies representative content within each frame, and optimal transport (OT) is then solved between neighboring frames to estimate temporal compressibility. We formulate this OT with non-uniform token mass, which protects semantically important tokens from aggressive compression, and a locality-aware cost that captures both feature and spatial disparities. The resulting transport plan jointly balances token importance and matching cost, while its total cost defines the transport difficulty of each frame pair, which we use to allocate compression budgets dynamically. Experiments on six benchmarks spanning video question answering and temporal grounding show that OTT-Vid preserves 95.8% of VQA and 73.9% of VTG performance while retaining only 10% of tokens, consistently outperforming existing state-of-the-art training-free compression methods.