Loading...
Loading...
Image-text pairs, instruction tuning, visual QA, cross-modal data, foundation model training data
1,956 datasets
ViTextVQA contains over 16,000 images and 50,000 question-answer pairs focused on Vietnamese text comprehension within visual contexts. Developed by researcher minhquan6203 and documented in Arxiv paper 2404.10652, it serves as a benchmark for text-based visual question answering in the Vietnamese language.
EyeVLM Dataset is a collection of data likely designed for training or evaluating vision-language models. The dataset is hosted on Kaggle, but its specific contents, size, and creation details are not provided in the available metadata. Further verification is required to confirm the exact nature and scope of the included data.
Kaggle hosts the SmolVLM_vigor_annotations dataset. The title suggests it contains annotations for evaluating vision-language models, likely on tasks like visual grounding or reasoning. The dataset's specific content, size, and origin require verification after download.
Ecological environment interaction data is collected in this multimodal dataset. The author, organization, and specific volume of data are not specified. The last update date is also unknown.
A multimodal dataset containing paired RGB and thermal image traces for reconstructing events in dynamic environments. The dataset is designed for research in multimodal sensor fusion and computer vision. Information on the creator, size, and specific temporal coverage is not provided in the input.
OLIMP is a heterogeneous multimodal dataset designed for advanced environment perception tasks. The dataset likely contains multiple data types, such as images, video, or sensor readings, integrated for perception modeling. Its author, organization, and specific size are not provided in the metadata.
RSCoVLM is a dataset for co-training vision-language models on remote sensing imagery for multi-task learning. The dataset is associated with a published academic paper and was created by a team of researchers including Qingyun Li, Shuran Ma, and Junwei Luo. It was last updated in January 2026.
AgentNet contains 22.6K human-annotated computer-use trajectories across Windows, macOS, and Ubuntu operating systems. Developed by xlangai and released in early 2026, it serves as a foundation for training vision-language-action (VLA) models for desktop automation.
Food-Health-VLM-Checkpoint is a dataset published on Kaggle. The title suggests it contains a checkpoint for a Vision-Language Model (VLM) trained on food and health-related data. Specific details regarding data volume, creation method, and authorship are not provided in the available metadata.
Food-health-vlm-checkpoint-best-1 is a model checkpoint likely containing trained weights for a Vision-Language Model (VLM). The dataset is hosted on Kaggle, but its specific contents, creation date, and author are not detailed in the provided metadata. Its title suggests a focus on applying multimodal AI to topics at the intersection of food and health.
A dataset titled 'WAVLM_Region(VF)' is hosted on Kaggle. The title suggests it contains audio feature representations, likely derived from the WAVLM model. No further metadata on size, format, or creation details is available.
Kaggle hosts the WAVLM_Gender(VF) dataset. The title suggests it contains speech audio data likely intended for gender classification tasks. Specific details on volume, creator, and creation date are unavailable.
A multimodal dataset related to Tamil culture and backdoor attacks in machine learning. The dataset is hosted on Kaggle, but its specific contents, size, and creation details are not provided in the available metadata. Further details regarding the data's origin, collection method, and temporal scope are unknown.
A dataset focused on emotion recognition, likely containing multiple data modalities such as text, audio, or images. It is hosted on Kaggle, a platform for data science and machine learning projects. The specific collection method, author, and temporal coverage are not detailed in the available metadata.
A multimodal dataset related to Tamil culture and backdoor attacks in machine learning. The dataset is hosted on Kaggle, but its specific contents, size, and creation details are not provided in the available metadata. Further details regarding the data's origin, collection method, and temporal scope are unknown.
EgoGazeVQA is an egocentric gaze-guided video question answering benchmark introduced in the paper 'In the Eye of MLLM: Benchmarking Egocentric Video Intent Understanding with Gaze-Guided Prompting'. The dataset, created by author taiyi09, leverages gaze information to improve the understanding of daily-life videos. It was last updated on 2026-01-22.
OctoCodingBench is a benchmark for evaluating scaffold-aware instruction following in repository-grounded agentic coding. It was created by MiniMaxAI and last updated on Hugging Face on January 13, 2026. The benchmark focuses on whether coding agents follow rules while solving tasks, a dimension not covered by existing task-completion benchmarks.
A dataset from Kaggle containing baseline results for the VIGOR-LLaVA-CAS benchmark. The dataset likely contains performance metrics and outputs for evaluating vision-language models. Its specific size, update date, and author are unknown.
NVIDIA's Nemotron-RL-instruction_following-structured_outputs dataset tests a model's ability to follow output formatting instructions under JSON schema constraints. Each problem consists of a document, an output formatting instruction (schema), and a question, with difficulty varied by instruction location, comprehensiveness, and schema complexity. The dataset was last updated on January 12, -2026.
TRACE is a multimodal dataset published on Kaggle. The dataset's specific content, size, and creation details are not provided in the available metadata. Further verification is required to determine its exact composition, scale, and origin.