Loading...
Loading...
Image-text pairs, instruction tuning, visual QA, cross-modal data, foundation model training data
1,947 datasets
SARLANG-1M is a large-scale benchmark for multimodal synthetic aperture radar image understanding. It comprises more than 1 million high-quality SAR image-text pairs collected from over 59 cities worldwide. The dataset was created by YiminJimmy and was last updated on February 10, 2026.
51,000 Kazakh question-answer pairs designed for instruction tuning of large language models. The dataset covers more than 20 domains, making it suitable for building general-purpose Kazakh language AI assistants. It is hosted on Kaggle and is formatted for immediate use in fine-tuning tasks.
Drishti-VLM-Data is a dataset published on Kaggle. The title suggests it contains data for training or evaluating vision-language models. The dataset's specific content, size, and origin are not detailed in the available metadata.
ViFoodVQA is a benchmark dataset for visual question answering tasks. The dataset likely contains images of Vietnamese food paired with questions and answers. It is hosted on Kaggle, but details about its size, creation, and update history are unknown.
A dataset associated with the ANN-2026 multimodal challenge, likely containing information related to crowdfunding campaigns and their outcomes. The dataset is hosted on Kaggle, but its specific contents, scale, and origin are not detailed in the available metadata. Further inspection after download is required to confirm the data's structure and features.
A balanced image-level tampering detection dataset in SmolVLM-style format. It includes chosen and rejected answer pairs derived from the pico-banana MCQ pipeline, suitable for preference learning and RLHF-style training. The dataset was created by author vanloc1808 and was last updated on February 15, 2026.
KitaKo Multimodal Dataset contains 110,000 images paired with 548,000 parallel captions. The captions are provided in three languages: English, Filipino, and Taglish. The dataset's author, organization, and last update date are unknown.
A dataset associated with the ICCV 2025 paper 'SENTINEL: Mitigating Object Hallucinations via Sentence-Level Early Intervention'. The dataset was created by author psp-dada and last updated on February 11, 2026. It is designed to address the problem of fabricated content in multimodal large language models.
A finetuned version of the BLIP model on the COCO dataset, likely containing image-text pairs for action captioning tasks. The dataset is hosted on Kaggle, but its specific size, columns, and creation details are unknown. Its content and scale require verification after download.
A dataset titled 'internvl24b-vlm-cia' is hosted on Kaggle. The name suggests it is likely a multimodal dataset for training or evaluating vision-language models. No further metadata is available to confirm its size, origin, or specific content.
Resized-Kvsair VQA is a dataset for visual question answering tasks, likely containing pairs of images and corresponding questions. It is hosted on Kaggle, a popular platform for data science competitions and datasets. The dataset's specific content, size, and creation details are not provided in the available metadata.
VLM Data is a dataset hosted on Hugging Face by author INV-WZQ. The dataset was last updated on April 1, 2026. Its specific content and scale are not detailed in the available metadata.
137,000 images containing Vietnamese text paired with 822,679 synthetic visual question-answering pairs generated by Gemini 1.5 Flash. Created by 5CD-AI and updated in February 2026, this collection focuses on Vietnamese OCR and scene understanding.
A finetuned version of the BLIP model, likely adapted for vision-language tasks. The dataset is hosted on Kaggle, but its specific content and scale are not detailed in the provided metadata. The original Flickr8K dataset is a standard benchmark for image captioning, suggesting this resource may contain model weights or related training data.
Boss Alignment Dataset is a collection for calibrating expectations of AI capabilities, likely containing examples or feedback. Authored by ChenZiHong-Gavin, it was last updated on GitHub on 2026-04-19. The specific content, scale, and structure require verification after download.
500 TEST VQA is a dataset for evaluating visual question answering models. It was published on Kaggle, but its author, organization, and creation date are unknown. The dataset's exact size, format, and annotation details require verification after download.
Multimodal Smart Grid Condition Records likely capture sensor data related to electrical relay performance. The dataset is hosted on Kaggle, but its specific size, origin, and update history are unspecified. Columns and sample data are unknown, requiring verification after download.
EgoBench is a multimodal interactive benchmark designed for evaluating tool-using agents. The benchmark likely contains tasks requiring agents to process and interact with multiple data modalities. Its specific size, format, and creation details are unknown.
VLMNS6 is a dataset published on Kaggle, a platform for data science competitions and open data. Its title suggests a focus on vision-language models, which combine computer vision and natural language processing. The dataset's specific content, scale, and origin are not detailed in the available metadata.
Sequential Movie Preference Dataset is a collection of user behavior data for personalized movie insights, published on Kaggle. The dataset likely contains sequences of user interactions or preferences related to movies. Its specific size, origin, and update history are not detailed in the provided metadata.