Research map / Vision-language and multimodal / Visual question answering
Learning to Count Objects in Natural Images for Visual Question Answering
Yan Zhang, Jonathon Hare, Adam Prügel-Bennett
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 differentiable counting from object proposals to the VQA baseline | VQA baseline · VQA v2 test-dev | Number accuracy: 46.88 -> 51.62 | Improved |
| Add differentiable counting from object proposals to the VQA baseline | VQA baseline · VQA v2 test | Number accuracy: 46.60 -> 51.39 | Improved |
| Add differentiable counting from object proposals to the VQA baseline | VQA baseline · VQA v2 validation | Number accuracy: 44.83$\pm$0.2 -> 49.36$\pm$0.1 | Improved |
| Add differentiable counting from object proposals to the VQA baseline | VQA baseline · VQA v2 validation, how-many questions | Count accuracy: 51.69$\pm$0.2 -> 57.03$\pm$0.0 | Improved |
| Add differentiable counting from object proposals to the VQA baseline | VQA baseline · VQA v2 validation | Balanced pair accuracy, Number: 17.34$\pm$0.2 -> 23.10$\pm$0.2 | Improved |
| Add differentiable counting from object proposals to the VQA baseline | VQA baseline · VQA v2 validation, how-many questions | Balanced pair accuracy, Count: 20.02$\pm$0.2 -> 26.63$\pm$0.2 | Improved |
| Replace the differentiable counting module with non-maximum suppression (ablation) | VQA baseline · VQA v2 validation | Number accuracy: 44.83$\pm$0.2 -> 44.60$\pm$0.1 | Hurt |
| Replace the differentiable counting module with non-maximum suppression (ablation) | VQA baseline · VQA v2 validation, how-many questions | Count accuracy: 51.69$\pm$0.2 -> 51.41$\pm$0.1 | Hurt |
| Replace the differentiable counting module with non-maximum suppression (ablation) | VQA baseline · VQA v2 validation | Balanced pair accuracy, Number: 17.34$\pm$0.2 -> 17.06$\pm$0.1 | Hurt |
| Replace the differentiable counting module with non-maximum suppression (ablation) | VQA baseline · VQA v2 validation, how-many questions | Balanced pair accuracy, Count: 20.02$\pm$0.2 -> 19.72$\pm$0.1 | Hurt |
Closest papers by meaning
- Interpretable Counting for Visual Question Answering (ICLR 2018 · similarity 0.94)
- An Analysis of Visual Question Answering Algorithms (ICCV 2017 · similarity 0.92)
- OK-VQA: A Visual Question Answering Benchmark Requiring External Knowledge (CVPR 2019 · similarity 0.91)
- Structured Attentions for Visual Question Answering (ICCV 2017 · similarity 0.91)
- Re-Attention for Visual Question Answering (AAAI 2020 · similarity 0.91)
- MoVie: Revisiting Modulated Convolutions for Visual Counting and Beyond (ICLR 2021 · 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.
