ARTeFACT benchmark for analog media defect segmentation
A collection of images and masks for evaluating analog media damage segmentation models.
- Image → image
- Image segmentation
- Image → text
Hugging Face datasets with plain-language descriptions: what's inside and what models to train on them. 4,788 in the catalog so far, with new ones added every hour.
Answered by Gemini 4 · the chat is kept only in this tab
A collection of images and masks for evaluating analog media damage segmentation models.
Large-scale English dataset of tens of millions of medical image-caption pairs from PubMed articles.
Dataset of 100 pages with fine-grained layout annotations, HTML tables, and text.
Official benchmark from Google DeepMind and Stanford for evaluating MLLM visual reasoning in polar coordinates.
A dataset of synthetic printed forms filled in by hand for text recognition tasks.
A set of 100 synthetic business documents with colored stamps and seals for computer vision tasks.
Multimodal text data in Indian languages for infrastructure auditing using VLM.
A set of images with Chinese text (100K–1M examples) from the FudanVI benchmark for training OCR systems.
120 corporate document images with various degradation levels for OCR testing.
A massive corpus of 57.3M short and long English captions for 10M Flickr images from various VLM models.
An SQLite database of Danbooru2023 tags and posts with a latest ID of 7,866,491 for generative models.
Manually verified image and transcription pairs of handwritten Vietnamese lines for OCR tasks.
Over 52k cropped text fragments from English scanned receipts for training OCR models.
Documents with text at various angles for OCR and recognition testing.
About 10k images of handwritten German text with transcripts for training recognition systems.
100K synthetic image-text pairs about anime characters and celebrities for training RAG-equipped VLMs.
Over 130,000 examples of Oogiri humor games with images and user ratings in three languages.
60 high-resolution pages with small text to test OCR robustness against complex structures.
Images of student handwritten math answers with teacher descriptions and QA pairs.
A large open multimodal dataset for training Vietnamese visual question-answering models.
Scanned old Buddhist texts in Sinhala and Pali paired with proofread transcriptions for OCR.
8.56M filtered English text-image pairs from LAION-COCO with aesthetic and watermark scores.
Large Stable Diffusion image corpus with user prompts and generation parameters.
Filtered ArtBench subset with paintings at 256x256 resolution for image generation and captioning.