Research map / Vision-language and multimodal / Visual question answering
Simple is not Easy: A Simple Strong Baseline for TextVQA and TextCaps
Qi Zhu, Chenyu Gao, Peng Wang, Qi Wu
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 |
|---|---|---|---|
| Separate OCR features into visual and linguistic branches rather than using one shared OCR attention block (ablation) | Simple TextVQA baseline · TextVQA val | accuracy: 1.77% | Improved |
| Add a global-feature attention block (ablation) | Two-block TextVQA baseline · TextVQA val | accuracy: 0.24% | Improved |
| Add an object-feature attention block (ablation) | Two-block TextVQA baseline · TextVQA val | accuracy: 0.63% | Improved |
| Add Recog-CNN OCR features (ablation) | Three-block TextVQA baseline · TextVQA val | accuracy: 0.37% | Improved |
| Increase Transformer depth from 4 to 8 layers (ablation) | Three-block TextVQA baseline · TextVQA val | accuracy: 0.1% | Improved |
| Replace Rosetta-en with SBD-Trans OCR (ablation) | Three-block TextVQA baseline · TextVQA val | accuracy: from $40.38\%$ to $43.95\% | Improved |
| Add ST-VQA training data (ablation) | Three-block TextVQA baseline · TextVQA val | accuracy: 45.53 | Improved |
| Add the policy-gradient auxiliary loss (ablation) | TextVQA model trained on ST-VQA · ST-VQA | performance: roughly $1\%$ improvement | Improved |
Closest papers by meaning
- Structured Multimodal Attentions for TextVQA (TPAMI 2021 · similarity 0.92)
- BLIVA: A Simple Multimodal LLM for Better Handling of Text-Rich Visual Questions (AAAI 2024 · similarity 0.89)
- VCR: A Task for Pixel-Level Complex Reasoning in Vision Language Models via Restoring Occluded Text (ICLR 2025 · similarity 0.88)
- PP-OCRv5: A Specialized 5M-Parameter Model Rivaling Billion-Parameter Vision-Language Models on OCR Tasks (CVPR 2026 · similarity 0.88)
- OCRBench v2: An Improved Benchmark for Evaluating Large Multimodal Models on Visual Text Localization and Reasoning (NeurIPS 2025 · similarity 0.88)
- Mining Fine-Grained Image-Text Alignment for Zero-Shot Captioning via Text-Only Training (AAAI 2024 · 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.
