Research map / Video / Video-language understanding
VideoLoom: A Video Large Language Model for Joint Spatial-Temporal Understanding
Jiapeng Shi, Junke Wang, Zuyao You, Bo He, Zuxuan Wu
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 |
|---|---|---|---|
| use SlowFast tokens with joint training instead of slow-only or fast-only tokens (ablation) | VideoLoom (InternVL2.5-4B) · QVHighlights / MeVIS | mIoU / J&F: +4.8 mIoU on QVHighlights, +3.2 J&F on MeVIS over standalone models | Improved |
| joint training with slow tokens only (no fast tokens) (ablation) | VideoLoom (InternVL2.5-4B) · Charades-STA | R1@0.5: 66.1 -> 38.8 (vs temporal-only fast) | Hurt |
| joint training with fast tokens only (no slow tokens) (ablation) | VideoLoom (InternVL2.5-4B) · MeVIS | J&F: 50.0 -> 44.6 | Hurt |
| split tokens by task (fast for temporal, slow for spatial) instead of unified SlowFast sequence (ablation) | VideoLoom (InternVL2.5-4B) · QVHighlights / MeVIS | mIoU / J&F: QVHighlights mIoU 57.2 -> 47.1; MeVIS 50.0 -> 47.6 | Hurt |
| add LoomData-8.7k to training data (ablation) | VideoLoom (InternVL2.5-4B) · LoomBench Combined | J&F bi-fore: 36.9 -> 41.9 (+5.0) | Improved |
| add LoomData-8.7k to training data (ablation) | VideoLoom (InternVL2.5-4B) · Charades-STA / ReVOS / MME / MMBench | R1@0.5 / J&F / MME / Acc: R1@0.5 66.3 -> 67.8; ReVOS 59.9 -> 60.6; MMBench 80.9 -> 81.1 | Improved |
| add LoomData-8.7k to training data (ablation) | VideoLoom · RefDavis17 (zero-shot, non-human) | J&F: 67.7 -> 70.0 (+2.3) | Improved |
| add LoomData-8.7k to training data (ablation) | VideoLoom · MeVIS interaction queries (275) | J&F: 63.6 -> 65.2 (+1.6) | Improved |
| add VQA data alone without LoomData (ablation) | VideoLoom (InternVL2.5-4B) · Table 6 benchmarks | ReVOS J&F / LoomBench J&F bi-fore: ReVOS 60.0 -> 59.9; LoomBench J&F bi-fore 34.6 -> 36.9 | No clear effect |
| scale backbone from InternVL2.5-4B to InternVL2.5-8B (ablation) | VideoLoom · LoomBench Combined | J&F bi-fore: 41.9 -> 47.2 | Improved |
| upgrade backbone from InternVL2.5-8B to InternVL3-8B (ablation) | VideoLoom · LoomBench Combined | J&F bi-fore: 47.2 -> 49.1 | Improved |
| joint model instead of cascaded TimeSuite+Sa2VA pipeline | VideoLoom-8B · LoomBench Combined | tIoU / J&F bi-fore: +16.2 tIoU, +15.4 J&F bi-fore | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 3 of 3 on most shared results.
- Sa2VA-8B — Sa2VA: Marrying SAM2 With MLLM for Dense Grounded Understanding of Images and Videos (TPAMI 2026) · better on 4 of 4 shared results
- GLUS-7B — GLUS: Global-Local Reasoning Unified into A Single Large Language Model for Video Segmentation (CVPR 2025) · better on 1 of 1 shared result
- TRACE-7B — TRACE: Temporal Grounding Video LLM via Causal Event Modeling (ICLR 2025) · 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
- LLaVA-ST: A Multimodal Large Language Model for Fine-Grained Spatial-Temporal Understanding (CVPR 2025 · similarity 0.90)
- VideoRefer Suite: Advancing Spatial-Temporal Object Understanding with Video LLM (CVPR 2025 · similarity 0.90)
- SAMA: Towards Multi-Turn Referential Grounded Video Chat with Large Language Models (NeurIPS 2025 · similarity 0.89)
- LongVideoBench: A Benchmark for Long-context Interleaved Video-Language Understanding (NeurIPS 2024 · similarity 0.89)
- Empowering LLMs with Pseudo-Untrimmed Videos for Audio-Visual Temporal Understanding (AAAI 2025 · similarity 0.89)
- Unleashing Hour-Scale Video Training for Long Video-Language Understanding (NeurIPS 2025 · similarity 0.89)
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.
