Research map / Video / Video-language understanding
DisTime: Distribution-based Time Representation for Video Large Language Models
Yingsen Zeng, Zepeng Huang, Yujie Zhong, Chengjian Feng, Jie Hu, Lin Ma, Yang Liu
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 changed | Where | Measured | Effect |
|---|---|---|---|
| predict timestamps via a distribution decoder instead of direct regression (ablation) | DisTime (InternVL2.5-1B) · Charades-STA | R@1 IoU=0.5: 51.9 -> 53.5 | Improved |
| predict timestamps via a distribution decoder instead of direct regression (ablation) | DisTime (InternVL2.5-1B) · YouCook2 | F1 Score: 2.2 -> 16.3 | Improved |
| add time token re-encoding (ablation) | DisTime (InternVL2.5-1B) · Charades-STA | R@1 IoU=0.7: 26.7 -> 29.7 | Improved |
| add time token re-encoding (ablation) | DisTime (InternVL2.5-1B) · YouCook2 | CIDEr: 4.0 -> 15.6 | Improved |
| add InternVid-TG training data (ablation) | DisTime (InternVL2.5-1B) · Charades-STA | R@1 IoU=0.7: 22.9 -> 29.7 | Improved |
| add InternVid-TG training data (ablation) | DisTime (InternVL2.5-1B) · QVHighlights | mAP@.5: 34.8 -> 47.9 | Improved |
| add Momentor data to training (ablation) | DisTime (InternVL2.5-1B) · Charades-STA | R@1 IoU=0.7: 22.9 -> 21.6 | Hurt |
| expand InternVid-TG from 90k-video subset to full 179k videos (ablation) | DisTime (InternVL2.5-1B) · Charades-STA | R@1 IoU=0.5: 55.5 -> 56.3 | Improved |
| use 32 bins instead of 64 in time decoder (ablation) | DisTime (InternVL2.5-1B) · Charades-STA | R@1 IoU=0.5: 53.7 -> 56.3 | Improved |
| use 32 bins instead of 16 in time decoder (ablation) | DisTime (InternVL2.5-1B) · YouCook2 | CIDEr: 12.1 -> 15.6 | Mixed |
| use 3 layers in time encoder/decoder instead of 2 (ablation) | DisTime (InternVL2.5-1B) · Charades-STA | R@1 IoU=0.5: 52.4 -> 56.3 | Improved |
| use 4 layers instead of 3 in time encoder/decoder (ablation) | DisTime (InternVL2.5-1B) · Charades-STA | R@1 IoU=0.5: 56.3 -> 54.4 | Hurt |
Closest papers by meaning
- Omnipotent Distillation with LLMs for Weakly-Supervised Natural Language Video Localization: When Divergence Meets Consistency (AAAI 2024 · similarity 0.90)
- Empowering LLMs with Pseudo-Untrimmed Videos for Audio-Visual Temporal Understanding (AAAI 2025 · similarity 0.89)
- TimeChat: A Time-sensitive Multimodal Large Language Model for Long Video Understanding (CVPR 2024 · similarity 0.89)
- VTG-LLM: Integrating Timestamp Knowledge into Video LLMs for Enhanced Video Temporal Grounding (AAAI 2025 · similarity 0.88)
- Momentor: Advancing Video Large Language Model with Fine-Grained Temporal Reasoning (ICML 2024 · similarity 0.88)
- Enhancing Temporal Understanding in Video-LLMs through Stacked Temporal Attention in Vision Encoders (NeurIPS 2025 · similarity 0.88)
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.
