LLaVA-OneVision-2 Multimodal and Video Data
Large collection of video clips, captions, and spatial reasoning tasks for training multimodal models.
- Video text → text
- Visual question answering
- Image text → text
- English
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
Large collection of video clips, captions, and spatial reasoning tasks for training multimodal models.
Real-world robot failure logs with visual symbols for error diagnosis and correction.
A collection of AI-generated 3D assets with previews and source inputs, totaling around 39 GB.
A curated collection of 59 English datasets for training VLMs in logic, STEM, charts, and OCR.
1.2 million image-text pairs for seven editing tasks, filtered via GPT-4o scores.
Massive dataset of physically simulated scenes (100M+ examples) from NVIDIA Isaac Sim for world models and robotics.
Massive, regularly updated scrape of public videos, comments, and stats from the Chinese platform Bilibili.
Clash Royale gameplay screenshots at roughly 10 frames per second for computer vision and game analysis.
Synthetic images and texts for point tracking, 3D reconstruction, and robotics tasks.
A test dataset from a CVPR 2026 paper featuring handheld recordings of everyday objects for detection tasks.
Over a thousand verified USD 3D objects with textures and physics, created in collaboration with NVIDIA Robotics.
Largest collection of schematics and PCBs: sources, images, metadata, and related board files.
A multimodal dataset with 3D data, images, and videos of robot motions, including stair climbing and descending.
Classic 70,000 MNIST digit images augmented with model embeddings and predictions.
Well-known dataset for detecting defects in images (non-commercial use only).
Curated selection of high-quality 4K images with text descriptions for ultra-high-resolution synthesis.
A collection of generated handwritten signatures rendered using 30 different Google Fonts handwriting fonts.
Dataset and evaluations for checking the safety of multimodal agents on web and mobile tasks.
Live microscopy frame sequences with cell cycle stage annotations for temporal classification.
22M image-text-to-text instructions for training and fine-tuning large multimodal models.
A large-scale multimodal chart corpus with images and text for data extraction, VQA, and multilingual understanding.
Nearly 70,000 English and Chinese examples for training educational LLMs and VLMs.
Large English-Chinese corpus of images, texts, and documents for training multimodal models.
Unofficial reconstruction of the well-known RDD2022 dataset for road defect detection.