ForestPrune achieves up to 90% token reduction in video MLLMs with minimal accuracy loss using a training-free spatial-temporal forest modeling approach.
March 25, 2026
Original Paper
ForestPrune: High-ratio Visual Token Compression for Video Multimodal Large Language Models via Spatial-Temporal Forest Modeling
arXiv · 2603.22911
The Takeaway
Processing video frames is computationally expensive for MLLMs; this method provides a massive speedup without requiring retraining. By identifying and preserving 'token trees' across frames, it maintains global video comprehension while drastically reducing the input sequence length.
From the abstract
Due to the great saving of computation and memory overhead, token compression has become a research hot-spot for MLLMs and achieved remarkable progress in image-language tasks. However, for the video, existing methods still fall short of high-ratio token compression. We attribute this shortcoming to the insufficient modeling of temporal and continual video content, and propose a novel and training-free token pruning method for video MLLMs, termed ForestPrune, which achieves effective and high-ra