Loading...
Loading...
Image-text pairs, instruction tuning, visual QA, cross-modal data, foundation model training data
1,956 datasets
Surveillance video data supports anomaly and crime detection tasks. The dataset is tagged for applications in security monitoring and video analysis. Specific details on volume, features, and origin are unavailable.
A validation dataset for the Visual Question Answering (VQA) task, published on Kaggle. The dataset likely contains image-question-answer pairs designed to test models' ability to answer questions about visual content. The specific number of samples, data source, and creation date are not provided in the available metadata.
VIGOR_annotations_llava is a dataset published on Kaggle. Its title suggests it contains annotations for the LLaVA (Large Language and Vision Assistant) model framework, likely linking images with descriptive text. The specific content, scale, and origin require verification after download.
vigor_annotations_llava is a dataset hosted on Kaggle. The title suggests it contains annotations likely intended for training or evaluating vision-language models, such as those based on the LLaVA architecture. Specific details regarding the data volume, creation method, and update history are not provided in the available metadata.
Multimodal data likely collected for assessing cognitive load, a psychological state related to mental effort. The dataset is published on Kaggle, but its specific size, collection method, and authorship are unknown. Its content and structure require verification after download.
A dataset from Kaggle focused on rumors circulating on the social media platform Twitter. The dataset likely contains multimodal content, such as text and associated images or videos, for analysis. Metadata is minimal; actual content, size, and collection details require verification after download.
Nemotron-Instruction-Following-Chat-v1 is designed to strengthen model capabilities in open-ended chat, precise instruction following, and structured output generation. It combines refreshed multi-turn chat data with synthetic dialogues generated by frontier models like GPT-OSS-120B and Qwen3-235B variants. The dataset was created by NVIDIA and last updated on December 15, 2025.
A research dataset for self-supervised learning on multimodal time-series data. It is designed for contrastive and adversarial augmentation techniques. The dataset's origin, size, and specific temporal coverage are not detailed in the provided metadata.
Kaggle hosts a dataset for multimodal radiomic-genomic fusion via graph-augmented deep learning for early prediction. The dataset's specific content, size, and origin are unknown. It is categorized as research data.
A language model dataset titled 'colsmolvlm-instruct-500m-base', published on Kaggle. The title suggests it is likely related to instruction tuning for a 500 million parameter language model. The dataset's specific content, size, and authorship are not detailed in the provided metadata.
Vision and audio data streams are categorized into structural crack classes for concrete integrity assessment. Synchronized multimodal inputs pair visual surface evidence with acoustic signatures to facilitate structural health monitoring research.
Multimodal video and audio recordings categorized into single-label violence classes. This dataset provides synchronized visual and auditory data streams to support the development of automated violence detection models.
90 animal species are categorized within this Vietnamese-language Visual Question Answering (VQA) dataset. The collection pairs images of animals with corresponding Vietnamese text questions and answers to facilitate multimodal learning.
A large-scale collection of mathematical problems categorized for Multi-Model TIR tasks. It provides structured data for training and evaluating reasoning-based mathematical solvers through multi-step logic.
53,202 instruction-tuning examples covering over 200 specialized cybersecurity domains, including cloud-native threats and AI/ML security. Created by the Trendyol Security Team for training defensive security AI assistants, this dataset was expanded from an initial 21,000 rows. The dataset was last updated on December 16, 2025.
NautData is a large-scale underwater instruction-following dataset containing 1.45 million image-text pairs. It was constructed by H-EmbodVis to bridge the gap in large-scale underwater multi-task instruction-tuning datasets. The dataset was introduced in the paper NAUTILUS and is intended for advancing underwater scene understanding methods.
A sample subset of data for Visual Question Answering (VQA), a multimodal AI task. The dataset is hosted on Kaggle, but its specific size, origin, and update history are not detailed in the provided metadata. Content likely pairs images with corresponding questions and answer annotations.
753,715 medical image-text pairs totaling 49.37 GB, designed for fine-tuning models like LLaVA-Med++. The dataset, created by Kafoo and last updated in November 2025, is stored in JSONL format alongside its images. Its captions are notably concise, averaging 1.0 words in length.
A dataset from Kaggle, likely containing paired image and text data for training or evaluating vision-language models. The specific content, scale, and creation details are not provided in the available metadata.
A dataset titled 'multimodal-vlm2' hosted on Kaggle. The title suggests it contains data for training or evaluating Vision-Language Models, which typically integrate visual and textual information. The dataset's specific content, size, and origin are not detailed in the provided metadata.