One Trajectory, One Token: Grounded Video Tokenization via Panoptic Sub-object Trajectory

TL;DR

This paper introduces grounded video tokenization using object trajectories, significantly reducing tokens and computational costs while improving performance in video understanding tasks.

Contribution

The authors propose TrajViT, a novel video encoder that leverages panoptic sub-object trajectories for efficient and effective video tokenization, outperforming existing methods.

Findings

TrajViT outperforms ViT3D by 6% top-5 recall in video-text retrieval.

TrajViT achieves 5.2% better performance on VideoQA benchmarks.

TrajViT trains 4x faster and uses 18x less inference FLOPs than ViT3D.

Abstract

Effective video tokenization is critical for scaling transformer models for long videos. Current approaches tokenize videos using space-time patches, leading to excessive tokens and computational inefficiencies. The best token reduction strategies degrade performance and barely reduce the number of tokens when the camera moves. We introduce grounded video tokenization, a paradigm that organizes tokens based on panoptic sub-object trajectories rather than fixed patches. Our method aligns with fundamental perceptual principles, ensuring that tokenization reflects scene complexity rather than video duration. We propose TrajViT, a video encoder that extracts object trajectories and converts them into semantically meaningful tokens, significantly reducing redundancy while maintaining temporal coherence. Trained with contrastive learning, TrajViT significantly outperforms space-time ViT (ViT3D) across multiple video understanding benchmarks, e.g., TrajViT outperforms ViT3D by a large margin of 6% top-5 recall in average at video-text retrieval task with 10x token deduction. We also show TrajViT as a stronger model than ViT3D for being the video encoder for modern VideoLLM, obtaining an average of 5.2% performance improvement across 6 VideoQA benchmarks while having 4x faster training time and 18x less inference FLOPs. TrajViT is the first efficient encoder to consistently outperform ViT3D across diverse video analysis tasks, making it a robust and scalable solution.