Research map / Vision-language and multimodal / Visual question answering
Sentence Attention Blocks for Answer Grounding
Seyedalireza Khoshsirat, Chandra Kambhamettu
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 LayerNorm normalization layer to the cross-attention block (ablation) | Sentence Attention Block · VizWiz-VQA-Grounding val | Mean IoU: 70.1 -> 71.2 | Improved |
| replace sigmoid with softmax (ablation) | Sentence Attention Block · VizWiz-VQA-Grounding val | Mean IoU: 71.2 -> 72.7 | Improved |
| invert the bottleneck by expanding embeddings by a factor of two (ablation) | Sentence Attention Block · VizWiz-VQA-Grounding val | Mean IoU: 72.7 -> 73.5 | Improved |
| add higher-resolution scales to multi-scale fusion (1/32 to 1/32+1/16+1/8+1/4) (ablation) | Sentence Attention Block · VizWiz-VQA-Grounding val | Mean IoU: 71.4 -> 73.5 | Improved |
| use EfficientNet-B7 backbone instead of ResNet-50x3 (ablation) | Sentence Attention Block · VizWiz-VQA-Grounding val | Mean IoU: 71.9 -> 73.5 | Improved |
| use Swin Transformer V2-B instead of EfficientNet-B7 (ablation) | Sentence Attention Block · VizWiz-VQA-Grounding val | Mean IoU: 73.5 -> 72.8 | Hurt |
| use EfficientNetV2-L instead of EfficientNet-B7 (ablation) | Sentence Attention Block · VizWiz-VQA-Grounding val | Mean IoU: 73.5 -> 73.1 | Hurt |
| use MiniLMv2 sentence encoder instead of RoBERTa or MPNet (ablation) | Sentence Attention Block · VizWiz-VQA-Grounding val | Mean IoU: 72.8 / 73.1 -> 73.5 | Improved |
| apply Sentence Attention Block vs MAC-Caps | answer grounding · TextVQA-X test | Mean IoU: 26.1 -> 29.0 | Improved |
| apply Sentence Attention Block vs best prior method on VizWiz challenge | answer grounding · VizWiz-VQA-Grounding Challenge 2022 | Mean IoU: 70.6 -> 72.4 | Improved |
Closest papers by meaning
- Multi-Level Attention Networks for Visual Question Answering (CVPR 2017 · similarity 0.90)
- VQA Therapy: Exploring Answer Differences by Visually Grounding Answers (ICCV 2023 · similarity 0.90)
- Structured Attentions for Visual Question Answering (ICCV 2017 · 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 0.89)
- FOCUS: Internal MLLM Representations for Efficient Fine-Grained Visual Question Answering (NeurIPS 2025 · similarity 0.88)
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.
