Research map / Vision-language and multimodal / Visual question answering
Learning to Answer Visual Questions From Web Videos
Antoine Yang, Antoine Miech, Josef Šivic, Ivan Laptev, Cordelia Schmid
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 |
|---|---|---|---|
| pretrain on HowToVQA69M | VQA-T · iVQA (finetuned) | top-1 accuracy (%): 23.0 -> 35.4 | Improved |
| pretrain on HowToVQA69M | VQA-T · MSRVTT-QA (finetuned) | top-1 accuracy (%): 39.6 -> 41.5 | Improved |
| pretrain on HowToVQA69M | VQA-T · How2QA (finetuned) | top-1 accuracy (%): 80.8 -> 84.4 | Improved |
| pretrain on HowToVQA69M instead of HowTo100M | VQA-T · iVQA (zero-shot) | top-1 accuracy (%): 1.9 -> 12.2 | Improved |
| add visual modality (VQA-T vs QA-T) trained on HowToVQA69M | QA-T · iVQA (zero-shot) | top-1 accuracy (%): 4.4 -> 12.2 | Improved |
| use the proposed QA generation instead of Heilman et al. rule-based method | VQA-T · ActivityNet-QA (zero-shot) | top-1 accuracy (%): 1.1 -> 12.2 | Improved |
| use the proposed QA generation instead of Heilman et al. | VQA-T · iVQA (finetuned) | top-1 accuracy (%): 31.4 -> 35.4 | Improved |
| add MLM loss together with sampling without answer repetition (ablation) | VQA-T · iVQA (zero-shot) | top-1 accuracy (%): 11.1 -> 12.2 | Improved |
| add MLM loss alone (without answer-repetition-free sampling) (ablation) | VQA-T · iVQA (zero-shot) | top-1 accuracy (%): 11.1 -> 10.9 | Hurt |
| add sampling without answer repetition (ablation) | VQA-T · MSVD-QA (zero-shot) | top-1 accuracy (%): 6.4 -> 7.5 | Improved |
| scale pretraining data from 1% to 100% (ablation) | VQA-T · iVQA (finetuned) | top-1 accuracy (%): 24.2 -> 35.4 | Improved |
| add WebVidVQA3M pretraining on top of HowToVQA69M | VQA-T · MSVD-QA (finetuned) | top-1 accuracy (%): 46.3 -> 47.5 | Improved |
Closest papers by meaning
- Just Ask: Learning To Answer Questions From Millions of Narrated Videos (ICCV 2021 · similarity 0.98)
- Leveraging Video Descriptions to Learn Video Question Answering (AAAI 2017 · similarity 0.91)
- ActivityNet-QA: A Dataset for Understanding Complex Web Videos via Question Answering (AAAI 2019 · similarity 0.91)
- Zero-Shot Video Question Answering via Frozen Bidirectional Language Models (NeurIPS 2022 · similarity 0.91)
- Open-vocabulary Video Question Answering: A New Benchmark for Evaluating the Generalizability of Video Question Answering Models (ICCV 2023 · similarity 0.90)
- Commonsense Video Question Answering through Video-Grounded Entailment Tree Reasoning (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.
