MattBuildsafe/aip-clothing-grouping-full

这是一个视觉-语言SFT数据集，用于按服装相似性对安全摄像头拍摄的人物剪影进行分组。数据集包含1120个示例，这些示例来自仓库中的40个地址-摄像头-天组合。每个示例以OpenAI风格的聊天对话格式呈现，包含系统消息、用户消息（包含文本和图像URL）和助手消息（包含JSON响应）。数据集约束条件包括每个示例最多10张图像，图像大小限制为长边不超过1024像素，JPEG格式，base64内联，平均每个示例约195 KB。

Vision-language SFT dataset for grouping security-camera person cutouts by clothing similarity. 1120 examples drawn from 40 address-camera-day(s) in the warehouse. Each line is an OpenAI-style chat conversation containing system message, user message (with text and image URLs), and assistant message (with JSON response). Constraints include max 10 images per example, images <=1024px on long edge, JPEG, base64 inline, and ~195 KB per example on avg.