AI Paper Atlas

Research map / Video / Video-language understanding

IntentQA: Context-aware Video Intent Reasoning

Jiapeng Li, Ping Wei, Wenjuan Han, Lifeng Fan

ICCV 2023 · Video › Video-language understanding · approach: Video question answering · 35 citations (OpenAlex)

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 changedWhereMeasuredEffect
add GPT commonsense prior at test time (ablation)CaVIR · IntentQA test (Total)accuracy (%): 54.50 -> 57.64Improved
add triplet loader for contrastive samples (ablation)CaVIR · IntentQA test (Total)accuracy (%): 52.34 -> 53.80Improved
add VQL (video query language) (ablation)Base Model · IntentQA test (Total)accuracy (%): 51.36 -> 52.34Improved
add triplet margin loss (ablation)CaVIR · IntentQA test (Total)accuracy (%): 53.80 -> 54.50Improved
add VQL to full model (removing it) (ablation)CaVIR · IntentQA test (Total)accuracy (%): 57.08 -> 57.64Improved
add contrastive learning to full model (removing it) (ablation)CaVIR · IntentQA test (Total)accuracy (%): 55.72 -> 57.64Improved
randomly mask k nodes of cross-modal graph (ablation)CaVIR w/o GPT · IntentQA test (Total)accuracy (%): 54.5 -> 51.5Hurt
mask lowest-attention nodes of cross-modal graph (ablation)CaVIR w/o GPT · IntentQA test (Total)accuracy (%): 54.5 -> 54.05No clear effect
stricter contrastive sample criterion (t1 = 1 with exact action) versus looser (ablation)contrastive learning · IntentQA test (Total)accuracy (%): 52.67 -> 54.50 (best config)Improved
add 'Let's think step by step' to GPT prompt (ablation)GPT commonsense module · IntentQA test (Total)accuracy (%): prompt 1: 57.64 -> 58.43; prompt 2: 57.31 -> 57.83; prompt 3: 57.17 -> 57.36Mixed
pretrain on IntentQA with triplet loss then finetune on MSRVTT-QAVGT-based model · MSRVTT-QA testaccuracy (%): 39.00 -> 39.39Improved

Closest papers by meaning

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy