Research map / Video / Video-language understanding
Streaming VideoLLMs for Real-Time Procedural Video Understanding
Dibyadip Chatterjee, Edoardo Remelli, Yale Song, Bugra Tekin, Abhay Mittal, Bharat Bhatnagar, Necati Cihan Camgoz, Shreyas Hampali, Eric Sauser, Shugao Ma, Angela Yao, Fadime Sener
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 |
|---|---|---|---|
| add long-term verbalization and interleaving (ablation) | ProVideLLM-1B/5 · Ego4D GoalStep val | per-frame mAP: 12.1 -> 13.0 | Improved |
| add long-term verbalization and interleaving (ablation) | ProVideLLM-1B/5 · Ego4D GoalStep test | per-frame mAP: 12.2 -> 12.9 | Improved |
| add long-term verbalization (ablation) | ProVideLLM-8B/11+ · COIN next step forecasting | Top-1 accuracy: 52.1 -> 53.6 | Improved |
| use single-view ProVideLLM with verbalization vs TempAgg + Goal with 12 views | TempAgg + Goal · Assembly101 action anticipation | Top-5 recall: 12.0 -> 13.8 | Improved |
| replace 2-layer MLP with QFormer on DINOv2 (5 tokens) (ablation) | DINOv2 connector · EgoExo4D keystep val | accuracy: 32.4 -> 33.6 | Improved |
| replace QFormer with DETR-QFormer on DINOv2 (5 tokens) (ablation) | DINOv2 connector · EgoExo4D keystep val | accuracy: 33.6 -> 34.7 | Improved |
| replace QFormer with DETR-QFormer on DINOv2 (17 tokens) (ablation) | DINOv2 connector · EgoExo4D keystep val | accuracy: 38.1 -> 40.7 | Improved |
| replace QFormer with DETR-QFormer on SigLIP (5 tokens) (ablation) | SigLIP connector · EgoExo4D keystep val | accuracy: 32.9 -> 30.9 | Hurt |
| replace SigLIP with DINOv2 using DETR-QFormer (17 tokens) (ablation) | DETR-QFormer · EgoExo4D keystep val | accuracy: 33.6 -> 40.7 | Improved |
| use SigLIP CLS token instead of DINOv2 with 1 token MLP (ablation) | MLP connector · EgoExo4D keystep val | accuracy: 28.7 -> 31.8 | Improved |
| add interleaving to verbalization in the multimodal cache (ablation) | ProVideLLM-1B/5 · Ego4D GoalStep long video | runtime stability | Improved |
| add verbalization of long-term past instead of visual tokens only (ablation) | ProVideLLM-1B/5 · Ego4D GoalStep long video | memory: 40K -> 2.5K tokens | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 4 of 4 on most shared results.
- VEDIT — VEDIT: Latent Prediction Architecture For Procedural Video Representation Learning (ICLR 2025) · better on 2 of 2 shared results
- VideoLLM-online — VideoLLM-online: Online Video Large Language Model for Streaming Video (CVPR 2024) · better on 2 of 2 shared results
- EgoOnly — Ego-Only: Egocentric Action Detection without Exocentric Transferring (ICCV 2023) · better on 2 of 2 shared results
- EgoVLPv2 — EgoVLPv2: Egocentric Video-Language Pre-training with Fusion in the Backbone (ICCV 2023) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- VideoLLM-online: Online Video Large Language Model for Streaming Video (CVPR 2024 · similarity 0.86)
- StreamBridge: Turning Your Offline Video Large Language Model into a Proactive Streaming Assistant (NeurIPS 2025 · similarity 0.85)
- VideoLLM-MoD: Efficient Video-Language Streaming with Mixture-of-Depths Vision Computation (NeurIPS 2024 · similarity 0.85)
- StreamForest: Efficient Online Video Understanding with Persistent Event Memory (NeurIPS 2025 · similarity 0.85)
- Streaming Long Video Understanding with Large Language Models (NeurIPS 2024 · similarity 0.85)
- EchoPrune: Interpreting Redundancy as Temporal Echoes for Efficient VideoLLMs (NeurIPS 2026 · similarity 0.84)
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.
