TextOCR-GPT4o: Images with Multilingual Text Captioned by GPT-4o

TextOCR-GPT4o is Meta's TextOCR dataset captioned with emphasis on OCR using GPT-4o. The dataset is intended for generating benchmarks to compare Vision-Language Models to GPT-4o. CaptionEmporium published it on Hugging Face in June 2024.

Use Cases

• Benchmarking Vision-Language Models against GPT-4o based on the dataset's stated purpose
• OCR performance evaluation based on images containing text
• Multilingual text recognition in images based on the description of Spanish and Japanese text
• Generating synthetic captions for image-text datasets based on the platform tags

Strengths

• Captioned using GPT-4o, a state-of-the-art model
• Contains images with text in multiple languages, including Spanish and Japanese

Limitations

• Row count and dataset size are unknown, which may limit suitability assessment
• Column-level documentation is absent; field semantics must be inferred after download
• Access to images requires agreeing to Meta's terms of service

Provenance

Source

Meta's TextOCR dataset

Collection Method

Captioned with emphasis on OCR using GPT-4o

Freshness

Last updated 2024-06-13 21:33:13