Research map / Vision-language and multimodal / Visual question answering
VQA-LOL: Visual Question Answering under the Lens of Logic
Tejas Gokhale, Pratyay Banerjee, Chitta Baral, Yezhou Yang
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 logically composed training data (VQA-Compose) | LXMERT · VQA-Compose val | accuracy (%): 50.79 -> 85.03 | Improved |
| add logically composed training data (VQA-Compose) | LXMERT · VQA val | accuracy (%): 68.94 -> 67.85 | Hurt |
| add Supplement data to Compose training | LXMERT · VQA-Compose val | accuracy (%): 85.03 -> 70.28 | Hurt |
| replace LXMERT with LOL attention modules (qATT) (ablation) | LXMERT · VQA-Compose val | accuracy (%): 70.28 -> 79.88 | Improved |
| use full LOL (question + logic attention) instead of qATT only (ablation) | LOL · VQA-Compose val | accuracy (%): 79.88 -> 81.07 | Improved |
| add Frechet-Compatibility loss to LOL (Full) (ablation) | LOL (Full) · VQA-Compose val | accuracy (%): 81.07 -> 82.39 | Improved |
| add Frechet-Compatibility loss to LOL (Full) (ablation) | LOL (Full) · VQA val | accuracy (%): 68.86 -> 68.10 | Hurt |
| add Frechet-Compatibility loss to LXMERT (ablation) | LXMERT · VQA-Compose val | accuracy (%): 70.28 -> 75.31 | Improved |
| use LOL instead of LXMERT for generalization from single to multiple connectives | LOL · VQA-Compose val (multiple connectives) | accuracy (%): 61.99 -> 66.03 | Improved |
| use end-to-end LOL without a parser instead of LOL with trained parser (ablation) | LOL (full) · VQA-Compose + Supplement val overall | accuracy (%): 83.15 -> 85.10 | Improved |
| use trained parser vs oracle parser with LOL (ablation) | LOL (full) · VQA-Compose + Supplement val overall | accuracy (%): 87.15 -> 83.15 | Hurt |
| train LOL on closed questions only instead of LXMERT | LOL (lATT) · VQA-Supplement val (closed-question training) | accuracy (%): 88.15 -> 89.00 | Improved |
Closest papers by meaning
- Logical Implications for Visual Question Answering Consistency (CVPR 2023 · similarity 0.93)
- Compositional Substitutivity of Visual Reasoning for Visual Question Answering (ECCV 2024 · similarity 0.93)
- LoRA: A Logical Reasoning Augmented Dataset for Visual Question Answering (NeurIPS 2023 · similarity 0.92)
- Making the v in VQA Matter: Elevating the Role of Image Understanding in Visual Question Answering (CVPR 2017 · similarity 0.92)
- OK-VQA: A Visual Question Answering Benchmark Requiring External Knowledge (CVPR 2019 · similarity 0.92)
- When Open-Vocabulary Visual Question Answering Meets Causal Adapter: Benchmark and Approach (AAAI 2025 · similarity 0.92)
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.
