腾讯开源 WeVisOne-VL：统一目标级与像素级视觉感知
原标题：Paper page - WeVisOne-VL: Unified Visual Perception from Objects to Pixels
AI 摘要
腾讯（微信团队）发布 WeVisOne-VL（WeChat Vision All-in-One），一个通过结构化下一 token 预测统一目标级与像素级视觉感知的视觉语言模型。该模型借助特殊 <OBJ> token 将轻量像素解码器接入 VLM，支持检测、指代定位、计数、OCR、分割与抠图等任务，并采用三阶段课程式 SFT 加混合任务 RL 训练。在 30 多个基准上，该模型在目标级与像素级任务均取得领先表现，并保持对目标缺失查询的稳健拒绝能力，模型与代码已开源。
正文节选
WeVisOne-VL: Unified Visual Perception from Objects to Pixels Abstract WeVisOne-VL unifies object-level and pixel-level visual perception via structured token prediction, supporting detection, localization, counting, OCR, segmentation, and matting with a lightweight pixel decoder and multi-stage training. Existing vision-language models (VLMs) remain unreliable for fine-grained visual perception. In this report, we present WeVisOne-VL (WeChat Vision All-in-One), a powerful VLM that unifies objec