Research map / Vision-language and multimodal / Visual question answering
VQS: Linking Segmentations to Questions and Answers for Supervised Attention in VQA and Question-Focused Semantic Segmentation
Chuang Gan, Yandong Li, Haoxiang Li, Chen Sun, Boqing Gong
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 learned supervised attention features | MLP + ResNet · VQA Real Multiple Choice test-dev | accuracy: 67.4 -> 68.9 | Improved |
| add learned supervised attention features | MLP + Attri. · VQA Real Multiple Choice test-dev | accuracy: 68.4 -> 69.5 | Improved |
| replace ResNet image features with attribute features | MLP · VQA Real Multiple Choice test-dev | accuracy: 67.4 -> 68.4 | Improved |
| ensemble 10 models instead of single best | MLP + Attri. + Atten. · VQA Real Multiple Choice test-dev | accuracy: 69.5 -> 70.5 | Improved |
| supervise attention with VQS segmentations instead of human attention maps (ablation) | attention-augmented MLP · VQA Real MC test-dev (attention supervision comparison) | accuracy (All): 68.42 -> 68.94 | Improved |
| supervise attention with VQS segmentations instead of tight bounding boxes (ablation) | attention-augmented MLP · VQA Real MC test-dev (attention supervision comparison) | accuracy (All): 68.65 -> 68.94 | Improved |
| add attention supervised by HAT maps (ablation) | Plain MLP · VQA Real MC test-dev (attention supervision comparison) | accuracy (All): 67.49 -> 68.42 | Improved |
| increase attention mask resolution from 7x7 to 14x14 (ablation) | VQS attention · VQA Real MC test-dev (mask resolution) | accuracy (All): 67.41 -> 68.94 | Improved |
| increase attention mask resolution from 11x11 to 14x14 (ablation) | VQS attention · VQA Real MC test-dev (mask resolution) | accuracy (All): 68.36 -> 68.94 | Improved |
| use word2vec embedding instead of bag-of-words for question (ablation) | DeconvNet · QFSS test (14,875 questions), all types | IOU: 0.2687 -> 0.2979 | Improved |
| use LSTM embedding instead of bag-of-words for question (ablation) | DeconvNet · QFSS test (14,875 questions), all types | IOU: 0.2687 -> 0.3144 | Improved |
| use word2vec embedding instead of bag-of-words for question (ablation) | Mask aggregation · QFSS test (14,875 questions), all types | IOU: 0.3256 -> 0.3174 | Hurt |
Closest papers by meaning
- Making the v in VQA Matter: Elevating the Role of Image Understanding in Visual Question Answering (CVPR 2017 · similarity 0.92)
- A-OKVQA: A Benchmark for Visual Question Answering Using World Knowledge (ECCV 2022 · 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)
- OK-VQA: A Visual Question Answering Benchmark Requiring External Knowledge (CVPR 2019 · similarity 0.91)
- When Open-Vocabulary Visual Question Answering Meets Causal Adapter: Benchmark and Approach (AAAI 2025 · 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.
