AI Paper Atlas

Research map / Video / Video-language understanding

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

Chenhao Zheng, Jieyu Zhang, Mohammadreza Salehi, Ziqi Gao, Vishnu Iyengar, Norimasa Kobori, Quan Kong, Ranjay Krishna

ICCV 2025 · Video › Video-language understanding · approach: Long video understanding · 4 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
tokenize videos by panoptic sub-object trajectories instead of space-time patchesvideo encoder (ViT-Large) · zero-shot video-text retrievalR@5: ActivityNet txt2vid 35.3 -> 38.6Improved
use trajectory tokens instead of space-time patches on SSV2TrajViT · SSV2 action classificationtop-1 accuracy: 46.1 -> 45.8No clear effect
use TrajViT as VideoLLM encoder instead of ViT3DLlama3 VideoLLM · NextQA OEWUPS: 49.2 -> 65.1Improved
use TrajViT as VideoLLM encoder instead of ViT3DLlama3 VideoLLM · MLVUaccuracy: 19.2 -> 32.2Improved
add segmentation tracking to segmentation-only tokens (ablation)TrajViT · ablation 880k, K400K400 accuracy: 19.9 -> 25.9Improved
add segmentation to tracking-only tokens (ablation)TrajViT · ablation 880k, retrievalretrieval: 15.4 -> 17.8Improved
use DirectSAM instead of SAM for segmentation (ablation)TrajViT · ablation 880k, K400K400 accuracy: 23.5 -> 25.9Improved
use 1024 instead of 512 trajectory resolution (ablation)TrajViT · ablation 880kSSV2/K400/Retrieval: SSV2 22.5 vs 22.7; K400 26.3 vs 25.9No clear effect
use ResNet-18 instead of Hiera-small as feature extractor (ablation)TrajViT trajectory encoder · ablation 880k, K400K400 accuracy: 21 -> 25.9Improved
add explicit position branch (ablation)TrajViT trajectory encoder · ablation 880k, SSV2SSV2 accuracy: 20.6 -> 22.7Improved
use perceiver resampler instead of mean-pooling to aggregate frames (ablation)TrajViT trajectory encoder · ablation 880k, SSV2SSV2 accuracy: 19.2 -> 22.7Improved
increase query tokens per trajectory from 1 to 4 (ablation)TrajViT trajectory encoder · ablation 880k, K400K400 accuracy: 25.9 -> 26.2No clear effect

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy