Tag
Explore by tags
Provides curated short video clips (49- and 81-frame) with layered ground truth—edit layers, alpha mattes, and composite targets—for training and evaluating content-preserving layered diffusion video editing. Contains background-replace and object-add edits; Apache-2.0 licensed.
Generates persistent, explorable 3D worlds from a single image by synthesizing long-range, geometry-consistent video and reconstructing it into an explicit 3D Gaussian scene. Intended for internal research use under NVIDIA's research license.
Benchmarks document-parsing systems on real-world enterprise PDFs and images—evaluates tables, charts, content faithfulness, semantic formatting, and visual grounding with human-verified, rule-level tests. Ships with ~2,000 pages, ~169K test rules, and an open evaluation framework for end-to-end pipeline scoring.
Provides a ~9.2M-instance Japanese multimodal post-training dataset for vision–language models, combining image–text pairs, PDF corpora and generated VQA to improve Japanese VLM performance; access is restricted by Japanese copyright (download via llm-jp GitLab).
Provides paired before/after satellite images with question–answer annotations for semantic change understanding. Includes Yes/No and multiple-choice formats, delivered in Hugging Face datasets (streaming-friendly), suited for remote-sensing multimodal VQA and semantic change captioning research.
Drafts multiple tokens in parallel with a lightweight block-diffusion drafter to enable speculative decoding for faster LLM inference. Designed to pair with Qwen3.6-35B-A3B and reports up to ~2.9× throughput improvements on common benchmarks.
A 1.4M image–text style dataset for text-to-image generation and style transfer, produced by mapping 170K curated style prompts to 400K content prompts via Qwen-Image to yield strong intra-style consistency. Designed for training and evaluating style-aware generative models; license: other.
Provides a 30K+ problem multimodal, multilingual dataset of Olympiad-level math problems with expert solutions and a math-aware retrieval benchmark—includes images, hierarchical topics, provenance from official booklets, and LLM-assisted metadata (v0, CC BY 4.0).
High-resolution vision transformers pretrained on one billion human images for human-centric tasks such as pose estimation, body-part segmentation, surface-normal and pointmap prediction. Provides multiple backbone sizes and task-specific checkpoints; released under the Sapiens2 license.
Provides ~55K multimodal VQA items with matched contrastive pairs and model‑generated rationales across five categories (General, Reasoning, Math, Graph/Chart, OCR), enabling research on faithful visual reasoning and robustness. Train split: 54,844 examples; license unspecified—verify before use.
Curated multimodal training corpus for spatial intelligence: ~8.16M QA-style samples paired with ~2.72M unique images (≈1.1 TB). Provides JSONL annotations, a 1,000-sample preview, and 52 independent image archives — used to train SenseNova-SI models.
Provides ~85K contrastive visual question–answer pairs where each example contains an anchor and a matched counterpart (image, question, answer). Pairs span General, Reasoning, Math, Graph/Chart and OCR categories to help train and evaluate fine‑grained, faithful visual reasoning in VLMs.