Research map / Vision-language and multimodal / Cross-modal learning
Iterative Answer Prediction With Pointer-Augmented Multimodal Transformers for TextVQA
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 a multimodal transformer instead of pairwise fusion (ablation) | TextVQA model · TextVQA val | accuracy: 26.56 -> 29.55 | Helped |
| Add pretrained BERT question encoding (ablation) | M4C without decoder · TextVQA val | accuracy: 29.55 -> 30.15 | Helped |
| Use Rosetta-en instead of Rosetta-ml (ablation) | M4C without decoder · TextVQA val | accuracy: 30.15 -> 31.28 | Helped |
| Add OCR bounding-box features (ablation) | M4C without decoder · TextVQA val | accuracy: 31.28 -> 33.32 | Helped |
| Add Faster R-CNN OCR appearance features (ablation) | M4C without decoder · TextVQA val | accuracy: 33.32 -> 34.38 | Helped |
| Add PHOC character-level OCR features (ablation) | M4C without decoder · TextVQA val | accuracy: 34.38 -> 35.70 | Helped |
| Replace one-step classification with iterative answer decoding (ablation) | M4C · TextVQA val | accuracy: 35.70 -> 39.40 | Helped |
| Replace one-step classification with iterative answer decoding (ablation) | M4C · ST-VQA val | ANLS: 0.397 -> 0.472 | Helped |
| Replace one-step classification with iterative answer decoding (ablation) | M4C · OCR-VQA val | accuracy: 46.3 -> 63.5 | Helped |
| Add fixed answer vocabulary (ablation) | M4C with OCR copying · TextVQA val | accuracy: 31.76 -> 39.40 | Helped |
| Add dynamic OCR pointer copying (ablation) | M4C with fixed vocabulary · TextVQA val | accuracy: 14.94 -> 39.40 | Helped |
| Add ST-VQA as additional training data (ablation) | M4C · TextVQA test | accuracy: 39.01 -> 40.46 | Helped |
Closest papers by meaning
- When Open-Vocabulary Visual Question Answering Meets Causal Adapter: Benchmark and Approach (AAAI 2025 · similarity 0.88)
- Track the Answer: Extending TextVQA from Image to Video with Spatio-Temporal Clues (AAAI 2025 · similarity 0.88)
- BLIVA: A Simple Multimodal LLM for Better Handling of Text-Rich Visual Questions (AAAI 2024 · similarity 0.88)
- VTQA: Visual Text Question Answering via Entity Alignment and Cross-Media Reasoning (CVPR 2024 · similarity 0.88)
- FOCUS: Internal MLLM Representations for Efficient Fine-Grained Visual Question Answering (NeurIPS 2025 · similarity 0.87)
- Core-to-Global Reasoning for Compositional Visual Question Answering (AAAI 2025 · similarity 0.87)
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.