Research map / Large language models / Language, safety and interpretability
ChartReader: A Unified Framework for Chart Derendering and Comprehension without Heuristic Rules
Zhi-Qi Cheng, Qi Dai, Alexander G. Hauptmann
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 rule-free chart component detection and extended embeddings with multi-task training | T5 · ChartQA (test) | relaxed accuracy: 41.04 -> 52.60 | Improved |
| add center-point detection module (ablation) | chart component detection · EC400K Line | chart-to-table score: 0.63 -> 0.97 | Improved |
| add grouping of centers and keypoints (ablation) | chart component detection · EC400K Bar | chart-to-table score: 0.77 -> 0.95 | Improved |
| add appearance embedding to input encoding (ablation) | TaPas · PlotQA-V1 | accuracy: 59.10 -> 74.20 | Improved |
| add chart component detection (CCD) to input (ablation) | TaPas · PlotQA-V1 | accuracy: 52.30 -> 74.20 | Improved |
| add location embedding to input encoding (ablation) | TaPas · PlotQA-V2 | accuracy: 51.20 -> 56.20 | Improved |
| add multi-task training on top of full pretraining mix (ablation) | T5 · ChartQA (test) | accuracy: 42.2 -> 52.6 | Improved |
| add chart-to-text data to the pretraining mix with multi-task (ablation) | T5 · ChartQA (test) | accuracy: 43.2 -> 52.6 | Improved |
| mix all datasets in multi-task training rather than sequential training (ablation) | T5 · ChartQA (test) | accuracy: 51.6 -> 52.6 | Improved |
| add CCD module (without multi-task) (ablation) | T5 pretrained on PQA, finetuned on CQA · ChartQA (test) | accuracy: 39.2 -> 39.6 | No clear effect |
| add CCD module (ablation) | T5 · training cost | TPU hours: 13 -> 28 | Hurt |
| replace rule-based derendering with keypoint detection | Chart-to-table · EC400K | GPU hours: 57h -> 22h | Improved |
Closest papers by meaning
- ChartNet: A Million-Scale, High-Quality Multimodal Dataset for Robust Chart Understanding (CVPR 2026 · similarity 0.91)
- ChartSketcher: Reasoning with Multimodal Feedback and Reflection for Chart Understanding (NeurIPS 2025 · similarity 0.91)
- EvoChart: A Benchmark and a Self-Training Approach Towards Real-World Chart Understanding (AAAI 2025 · similarity 0.90)
- Visual Self-Refine: A Pixel-Guided Paradigm for Accurate Chart Parsing (ICLR 2026 · similarity 0.89)
- ChartE$^{3}$: A Comprehensive Benchmark for End-to-End Chart Editing (ICML 2026 · similarity 0.88)
- ChartMoE: Mixture of Diversely Aligned Expert Connector for Chart Understanding (ICLR 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.
