Research map / Vision-language and multimodal / Visual question answering
Exploring Human-Like Attention Supervision in Visual Question Answering
Tingting Qiao, Jianfeng Dong, Duanqing Xu
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 HAN-generated human-like attention supervision to MLB with 2 glimpses | MLB · VQA v2.0 test-dev | overall accuracy: 61.84 -> 61.99 | Improved |
| Add HAN-generated human-like attention supervision to MLB with 1 glimpse | MLB · VQA v2.0 test-dev | overall accuracy: 61.55 -> 61.66 | Improved |
| Use GRU to fuse attention maps instead of averaging them, with 2 glimpses (ablation) | HAN · VQA-HAT validation | mean rank-correlation: 0.406 ± 0.001 -> 0.656 ± 0.002 | Improved |
| Use GRU to fuse attention maps instead of averaging them, with 3 glimpses (ablation) | HAN · VQA-HAT validation | mean rank-correlation: 0.411 ± 0.001 -> 0.668 ± 0.001 | Improved |
| Use 2 glimpses instead of 1 (ablation) | HAN · VQA-HAT validation | mean rank-correlation: 0.536 -> 0.656 | Improved |
| Use 3 glimpses instead of 2 (ablation) | HAN · VQA-HAT validation | mean rank-correlation: 0.656 -> 0.668 | Improved |
| Use 4 glimpses instead of 3 (ablation) | HAN · VQA-HAT validation | mean rank-correlation: 0.668 -> 0.553 | Hurt |
| Add human-like attention supervision with 2 glimpses | MLB · VQA v2.0 test-dev, Number questions | accuracy: 37.52 -> 37.94 | Improved |
Closest papers by meaning
- Explanation vs Attention: A Two-Player Game to Obtain Attention for VQA (AAAI 2020 · similarity 0.92)
- Hierarchical Question-Image Co-Attention for Visual Question Answering (NeurIPS 2016 · similarity 0.92)
- Re-Attention for Visual Question Answering (AAAI 2020 · similarity 0.91)
- An Analysis of Visual Question Answering Algorithms (ICCV 2017 · similarity 0.91)
- Multi-Level Attention Networks for Visual Question Answering (CVPR 2017 · similarity 0.91)
- High-Order Attention Models for Visual Question Answering (NeurIPS 2017 · similarity 0.91)
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.
