Research map / Video / Video-language understanding
Tem-Adapter: Adapting Image-Text Pretraining for Video Question Answer
Guangyi Chen, Xiao Liu, Guangrun Wang, Kun Zhang, Philip H.S. Torr, Xiao-Ping Zhang, Yansong Tang
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 language template converting QA pairs to declarative sentences (ablation) | Tem-Adapter · SUTD-TrafficQA | average accuracy: 43.9 -> 46.0 | Improved |
| add Semantic Aligner (ablation) | Tem-Adapter · SUTD-TrafficQA | average accuracy: 44.2 -> 46.0 | Improved |
| add Temporal Aligner (ablation) | Tem-Adapter · SUTD-TrafficQA | average accuracy: 43.6 -> 46.0 | Improved |
| add language-guided auto-regression loss (ablation) | Tem-Adapter · SUTD-TrafficQA | average accuracy: 44.9 -> 46.0 | Improved |
| add auto-regression loss (ablation) | Tem-Adapter · MSR-VTT-MC | accuracy: 92.6 -> 94.3 | Improved |
| add both Temporal and Semantic Aligners (ablation) | CLIP + template · MSR-VTT-MC | accuracy: 74.1 -> 94.3 | Improved |
| add language guidance to the autoregressive Temporal Aligner (ablation) | Temporal Aligner · SUTD-TrafficQA reconstruction | PSNR: 28.7 -> 29.1 | Improved |
| more trainable parameters (full vs partial finetuning; multi-layer vs single adapter) | CLIP adaptation · SUTD-TrafficQA vs MSR-VTT-MC | accuracy: SUTD: 41.7 vs 40.3; MSR-VTT-MC: 88.5 vs 89.0 | Mixed |
| vary Semantic/Temporal Aligner hidden size and layer counts (ablation) | Tem-Adapter · SUTD-TrafficQA | accuracy: 44.6 to 45.8 versus default 46.0 | No clear effect |
| adapt CLIP features with Tem-Adapter inside HCRN | CLIP+HCRN · SUTD-TrafficQA | accuracy: 41.9 -> 46.0 | Improved |
Closest papers by meaning
- Temporal Reasoning Transfer from Text to Video (ICLR 2025 · similarity 0.89)
- Video-Context Aligned Transformer for Video Question Answering (AAAI 2024 · similarity 0.89)
- LongVideoBench: A Benchmark for Long-context Interleaved Video-Language Understanding (NeurIPS 2024 · similarity 0.89)
- Can Video LLMs Refuse to Answer? Alignment for Answerability in Video Large Language Models (ICLR 2025 · similarity 0.88)
- Video-Panda: Parameter-efficient Alignment for Encoder-free Video-Language Models (CVPR 2025 · similarity 0.88)
- Language-aware Visual Semantic Distillation for Video Question Answering (CVPR 2024 · similarity 0.88)
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.
