Research map / Video / Video-language understanding
Enrich and Detect: Video Temporal Grounding with Multimodal LLMs
Shraman Pramanick, Effrosyni Mavroudi, Yale Song, Rama Chellappa, Lorenzo Torresani, Triantafyllos Afouras
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 query enrichment (without MIL) (ablation) | ED-VTG · Charades-STA STG zero-shot | mIoU: 31.0 -> 37.7 | Improved |
| add query enrichment (without MIL) (ablation) | ED-VTG · ActivityNet-Captions STG zero-shot | mIoU: 29.6 -> 33.4 | Improved |
| add multiple-instance-learning objective on top of enrich and detect (ablation) | ED-VTG · Charades-STA STG zero-shot | mIoU: 37.7 -> 40.2 | Improved |
| add multiple-instance-learning objective on top of enrich and detect (ablation) | ED-VTG · ActivityNet-Captions STG zero-shot | mIoU: 33.4 -> 35.2 | Improved |
| add enrichment as offline training preprocessing instead of online two-step enrichment (ablation) | ED-VTG · Charades-STA STG (FT w/o PT) | mIoU: 33.2 -> 33.4 | No clear effect |
| replace offline enrichment with online enrich & detect (ablation) | ED-VTG · Charades-STA STG (FT w/o PT) | mIoU: 33.4 -> 38.4 | Improved |
| add offline paraphrasing of queries with blind LLaMA (ablation) | ED-VTG · Charades-STA STG (FT w/o PT) | mIoU: 33.2 -> 32.7 | No clear effect |
| add offline enrichment without cropping to the annotated interval (ablation) | ED-VTG · Charades-STA STG (FT w/o PT) | mIoU: 33.2 -> 31.9 | Hurt |
| add interval decoder with L1 and gIoU losses to LM-only training (ablation) | ED-VTG · Charades-STA STG (FT w/o PT) | mIoU: 34.1 -> 38.4 | Improved |
| add gIoU loss to decoder with L1 (ablation) | ED-VTG · Charades-STA STG (FT w/o PT) | mIoU: 37.0 -> 38.4 | Improved |
| add VPG data to STG pre-training (ablation) | ED-VTG · Charades-STA STG zero-shot | mIoU: 37.0 -> 39.8 | Improved |
| add AG data to STG + VPG pre-training (ablation) | ED-VTG · Charades-STA STG zero-shot | mIoU: 39.8 -> 39.9 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- TimeChat — TimeChat: A Time-sensitive Multimodal Large Language Model for Long Video Understanding (CVPR 2024) · better on 5 of 5 shared results
- VTimeLLM — VTimeLLM: Empower LLM to Grasp Video Moments (CVPR 2024) · 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
- Universal Video Temporal Grounding with Generative Multi-modal Large Language Models (NeurIPS 2025 · similarity 0.92)
- Empower Words: DualGround for Structured Phrase and Sentence-Level Temporal Grounding (NeurIPS 2025 · similarity 0.92)
- Unleashing the Potential of Multimodal LLMs for Zero-Shot Spatio-Temporal Video Grounding (NeurIPS 2025 · similarity 0.91)
- VTG-LLM: Integrating Timestamp Knowledge into Video LLMs for Enhanced Video Temporal Grounding (AAAI 2025 · similarity 0.90)
- SnAG: Scalable and Accurate Video Grounding (CVPR 2024 · similarity 0.90)
- ReVisionLLM: Recursive Vision-Language Model for Temporal Grounding in Hour-Long Videos (CVPR 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.
