Research map / Video / Video-language understanding
Learning Streaming Video Representation via Multitask Training
Yibin Yan, Jilan Xu, Shangzhe Di, Yikun Liu, Yudi Shi, Qirui Chen, Zeqian Li, Yifei Huang, Weidi Xie
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 global and temporal tasks (ablation) | StreamFormer spatial-only · THUMOS-14 OAD (w/o flow) | mAP: 51.5 -> 69.1 | Improved |
| add spatial-level tasks to global+temporal (ablation) | StreamFormer · YouTube-VIS 2019 | AP: 41.9 -> 45.1 | Improved |
| add spatial-level tasks to global+temporal (ablation) | StreamFormer · THUMOS-14 OAD (w/o flow) | mAP: 69.1 -> 68.4 | Hurt |
| add spatial-level tasks to global+temporal (ablation) | StreamFormer · VideoMME | accuracy: 43.1 -> 45.0 | Improved |
| add multitask video training over SigLIP image encoder | StreamFormer · THUMOS-14 OAD (w/o flow) | mAP: 58.9 -> 68.4 | Improved |
| add temporal attention (causal) to image model (ablation) | StreamFormer · THUMOS-14 OAD (w/o flow, 10% data) | mAP: 61.7 -> 62.5 | Improved |
| use bidirectional instead of causal temporal attention (ablation) | StreamFormer · THUMOS-14 OAD (w/o flow, 10% data) | mAP: 62.5 -> 62.8 | Mixed |
| use multitask annotated data instead of WebVid-1M contrastive (ablation) | StreamFormer · THUMOS-14 OAD (w/o flow) | mAP: 57.2 -> 62.5 | Improved |
| increase multitask pre-training data from 0.1M to 1M (ablation) | StreamFormer · THUMOS-14 OAD (w/o flow) | mAP: 62.5 -> 68.4 | Improved |
| increase multitask pre-training data from 0.1M to 1M (ablation) | StreamFormer · MLVU | accuracy: 49.6 -> 48.4 | Hurt |
| add spatial-only tasks to SigLIP baseline (ablation) | StreamFormer spatial-only · VideoMME | accuracy: 40.8 -> 30.0 | Hurt |
| replace SigLIP with StreamFormer in CTVIS | CTVIS · YouTube-VIS 2019 | AP: 40.9 -> 45.1 | Improved |
Closest papers by meaning
- StreamForest: Efficient Online Video Understanding with Persistent Event Memory (NeurIPS 2025 · similarity 0.89)
- Streaming Long Video Understanding with Large Language Models (NeurIPS 2024 · similarity 0.88)
- Streaming Video Understanding and Multi-round Interaction with Memory-enhanced Knowledge (ICLR 2025 · similarity 0.87)
- SVBench: A Benchmark with Temporal Multi-Turn Dialogues for Streaming Video Understanding (ICLR 2025 · similarity 0.87)
- CoRDS: Coreset-Based Representative and Diverse Selection for Streaming Video Understanding (NeurIPS 2026 · similarity 0.86)
- Learning from Streaming Video with Orthogonal Gradients (CVPR 2025 · similarity 0.86)
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.
