multimodalart/test-dataset-order

MONET是一个大规模多模态数据集，专注于文本到图像、图像特征提取和零样本图像分类任务。数据集包含超过1亿但少于10亿个样本，以英语为主。它整合了图像、多种美学评分（如JasperAI和LAION改进版）、多种AI模型生成的标题（包括Florence-2-large、Gemini-2.5-flash-lite、InternVL-3-8b、原始标题和ShareGPT4V-7b）、分类器输出（如CLIP和YOLO）、检测结果（如人脸检测和YOLO检测）、嵌入向量（如CLIP、DINOv2、SSCD和VAE）、哈希值（MD5、感知哈希和SHA256）、图像尺寸、许可证信息、NSFW评分以及来源URL等丰富特征。该数据集旨在支持多模态研究，特别是图像生成、字幕生成和内容分析。

MONET is a large-scale multimodal dataset focused on text-to-image, image-feature-extraction, and zero-shot-image-classification tasks. It contains between 100 million and 1 billion samples, primarily in English. The dataset integrates images, multiple aesthetic scores (e.g., JasperAI and LAION-improved), captions generated by various AI models (including Florence-2-large, Gemini-2.5-flash-lite, InternVL-3-8b, original captions, and ShareGPT4V-7b), classifier outputs (e.g., CLIP and YOLO), detection results (e.g., face detection and YOLO detection), embedding vectors (e.g., CLIP, DINOv2, SSCD, and VAE), hash values (MD5, perceptual hash, and SHA256), image dimensions, license information, NSFW scores, and source URLs. It is designed to support multimodal research, particularly in image generation, captioning, and content analysis.