Research map / Vision-language and multimodal / Visual question answering
Explanation vs Attention: A Two-Player Game to Obtain Attention for VQA
Badri Patro, Anupriy, Vinay Namboodiri
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 pixel-wise adversarial attention using Grad-CAM supervision | SAN · HAT | RC: 0.2432 -> 0.3071 | Improved |
| Add pixel-wise adversarial attention using Grad-CAM supervision | SAN · VQA-v1 test | All accuracy: 58.7 -> 63.6 | Improved |
| Add PAAN attention supervision | MCB · VQA-v1 test | All accuracy: 64.2 -> 67.1 | Improved |
| Add PAAN attention supervision | MCB · VQA-v2 test | All accuracy: 64.0 -> 68.4 | Improved |
| Use pixel-wise local discrimination rather than global adversarial discrimination (ablation) | AAN · HAT | RC: 0.2896 -> 0.3071 | Improved |
| Replace adversarial distribution matching with MMD (ablation) | PAAN · HAT | RC: 0.3071 -> 0.2573 | Hurt |
| Replace adversarial distribution matching with CORAL (ablation) | PAAN · HAT | RC: 0.3071 -> 0.2563 | Hurt |
| Replace adversarial distribution matching with MSE (ablation) | PAAN · HAT | RC: 0.3071 -> 0.2681 | Hurt |
| Use RISE instead of Grad-CAM as explanation supervision (ablation) | PAAN · HAT | RC: 0.3071 -> 0.2754 | Hurt |
| Use random explanation masks with 7% overlap instead of Grad-CAM (ablation) | PAAN · HAT | RC: 0.3071 -> 0.1213 | Hurt |
| Increase random explanation-mask overlap from 7% to 20% (ablation) | PAAN · HAT | RC: 0.1213 -> 0.1746 | Improved |
| Use pixel-wise PAAN rather than global AAN (ablation) | SAN · VQA-v1 test-dev | All accuracy: 62.3 -> 63.6 | Improved |
Closest papers by meaning
- Exploring Human-Like Attention Supervision in Visual Question Answering (AAAI 2018 · similarity 0.92)
- Multi-Level Attention Networks for Visual Question Answering (CVPR 2017 · similarity 0.90)
- Re-Attention for Visual Question Answering (AAAI 2020 · similarity 0.90)
- An Analysis of Visual Question Answering Algorithms (ICCV 2017 · similarity 0.90)
- Hierarchical Question-Image Co-Attention for Visual Question Answering (NeurIPS 2016 · similarity 0.90)
- Structured Attentions for Visual Question Answering (ICCV 2017 · similarity 0.90)
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.
