vcr-org/

VCR-wiki-zh-hard

🏠 Paper | 👩🏻‍💻 GitHub | 🤗 Huggingface Datasets | 📏 Evaluation with lmms-eval This is the official Hugging Face dataset for VCR-Wiki, a dataset for the Visual Caption Restoration (VCR) task. VCR is designed to measure vision-language models' capability to accurately restore partially obscured texts using pixel-level hints within images. text-based processing becomes ineffective in VCR as accurate text restoration depends… See the full description on the dataset page: https://huggingface.co/datasets/vcr-org/VCR-wiki-zh-hard.