How to Teach AI to Understand Charts: Analyzing the New ChartNet Dataset from MIT
Researchers from MIT and IBM Research have introduced a multimodal dataset called ChartNet for training and testing models on chart analysis. The dataset includes 1.7 million synthetic examples covering 24 visualization types, with automatic verification and manually validated subsets. Fine-tuning open VLM models on ChartNet allowed models with 2–7 billion parameters to compete with large systems, occasionally surpassing GPT-4o.
IBM
Alibaba/Qwen
Google/DeepMind
OpenAI
Hugging Face
Habr — хаб NLP27.07 · 01:05