Research map / Video / Video-language understanding
TOGA: Temporally Grounded Open-Ended Video QA with Weak Supervision
Ayush Gupta, Anirban Roy, Rama Chellappa, Nathaniel D. Bastian, Alvaro Velasquez, Susmit Jha
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 final consistency-constraint stage (ablation) | TOGA · NExT-GQA | mIoU: 12.1 -> 24.4 | Improved |
| use multi-scale connector instead of sparse only (ablation) | TOGA · NExT-GQA | mIoU: 20.0 -> 24.4 | Improved |
| use multi-scale connector instead of dense only (ablation) | TOGA · NExT-GQA | mIoU: 22.1 -> 24.4 | Improved |
| multi-scale connector on short events (ablation) | TOGA · NExT-GQA short events | IoU: 18.3 -> 20.5 | Improved |
| multi-scale connector on long events (ablation) | TOGA · NExT-GQA long events | IoU: 32.1 -> 49.3 | Improved |
| multi-scale connector overall Acc@GQA (ablation) | TOGA · NExT-GQA | Acc@GQA: 20.12 -> 24.6 | Improved |
| use (16,4) frames instead of (8,2) (ablation) | TOGA · NExT-GQA | mIoU: 20.8 -> 24.4 | Improved |
| use (16,4) frames instead of (32,8) (ablation) | TOGA · NExT-GQA | mIoU: 21.5 -> 24.4 | Improved |
| use 4X sparse downsampling (16,4) instead of 2X (16,8) (ablation) | TOGA · NExT-GQA | mIoU: 21.7 -> 24.4 | Improved |
| use integer [0,100] timestamps instead of float [0,1] (ablation) | TOGA · NExT-GQA | mIoU: 19.0 -> 24.4 | Improved |
| add frame indices in the prompt (ablation) | TOGA · NExT-GQA | mIoU: 24.4 -> 21.2 | Hurt |
| use Mistral-7B instead of Vicuna-7B decoder (ablation) | TOGA · NExT-GQA | mIoU: 19.0 -> 24.4 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- VTimeLLM — VTimeLLM: Empower LLM to Grasp Video Moments (CVPR 2024) · better on 3 of 3 shared results
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
- Searching Videos as Trees: Self-Correcting Agents for Grounded Long Video QA (NeurIPS 2026 · similarity 0.88)
- Can I Trust Your Answer? Visually Grounded Video Question Answering (CVPR 2024 · similarity 0.88)
- Grounded Question-Answering in Long Egocentric Videos (CVPR 2024 · similarity 0.88)
- Universal Video Temporal Grounding with Generative Multi-modal Large Language Models (NeurIPS 2025 · similarity 0.87)
- Commonsense Video Question Answering through Video-Grounded Entailment Tree Reasoning (CVPR 2025 · similarity 0.87)
- STAIR: Spatial-Temporal Reasoning with Auditable Intermediate Results for Video Question Answering (AAAI 2024 · similarity 0.87)
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.
