Research map / Vision-language and multimodal / Visual question answering
Interpretable Counting for Visual Question Answering
Alexander Trott, Caiming Xiong, Richard Socher
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 |
|---|---|---|---|
| use sequential discrete object selection with learned object interactions instead of attention-based answer classification | IRLC vs UpDown · HowMany-QA test | accuracy: 52.7 -> 57.7 | Improved |
| use sequential discrete object selection with learned object interactions instead of attention-based answer classification | IRLC vs UpDown · HowMany-QA test | RMSE: 2.64 -> 2.37 | Improved |
| use sequential discrete object selection instead of fractional object counts | IRLC vs SoftCount · HowMany-QA test | accuracy: 50.2 -> 57.7 | Improved |
| use sequential discrete object selection instead of fractional object counts | IRLC vs SoftCount · HowMany-QA test | RMSE: 2.37 -> 2.37 | No clear effect |
| add caption grounding during training (ablation) | IRLC · HowMany-QA test | accuracy: 56.1 -> 57.7 | Improved |
| add caption grounding during training (ablation) | IRLC · HowMany-QA test | RMSE: 2.45 -> 2.37 | Improved |
| add caption grounding during training (ablation) | UpDown · HowMany-QA test | accuracy: 51.5 -> 52.7 | Improved |
| add caption grounding during training (ablation) | SoftCount · HowMany-QA test | accuracy: 49.2 -> 50.2 | Improved |
| add Visual Genome QA training data (ablation) | counting models · HowMany-QA test | accuracy: 2.7% | Improved |
| add Visual Genome QA training data (ablation) | counting models · HowMany-QA test | RMSE: 0.12 | Improved |
| include entropy and interaction auxiliary penalties during training (ablation) | IRLC · HowMany-QA development | accuracy | Improved |
| increase the weights of entropy and interaction auxiliary penalties too much (ablation) | IRLC · HowMany-QA development | accuracy | Hurt |
Closest papers by meaning
- Learning to Count Objects in Natural Images for Visual Question Answering (ICLR 2018 · similarity 0.94)
- An Analysis of Visual Question Answering Algorithms (ICCV 2017 · similarity 0.92)
- TallyQA: Answering Complex Counting Questions (AAAI 2019 · similarity 0.91)
- MoVie: Revisiting Modulated Convolutions for Visual Counting and Beyond (ICLR 2021 · similarity 0.91)
- Revisiting Visual Question Answering Baselines (ECCV 2016 · similarity 0.90)
- OK-VQA: A Visual Question Answering Benchmark Requiring External Knowledge (CVPR 2019 · 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.
