Vlm on HJJJ的个人小站https://hjjj.top/tags/vlm/Recent content in Vlm on HJJJ的个人小站Hugoen-usSun, 16 Aug 2026 11:11:11 +0800ViT 详解：如何把一张图变成一段 token 序列https://hjjj.top/posts/computer-science/vlm/vit-intro/Thu, 13 Aug 2026 21:30:00 +0800https://hjjj.top/posts/computer-science/vlm/vit-intro/从 Patch Embedding 的张量变形到图像分类：Vision Transformer 的原理、实现与局限LLaVA 详解：视觉指令微调的完整拆解https://hjjj.top/posts/computer-science/vlm/llava-intro/Thu, 13 Aug 2026 00:00:00 +0800https://hjjj.top/posts/computer-science/vlm/llava-intro/从线性投影到自回归交叉熵损失，完整推导 LLaVA 如何把图像变成语言模型能读的 tokenCLIP 详解：对比学习如何让模型同时看懂图与文https://hjjj.top/posts/computer-science/vlm/clip-intro/Wed, 12 Aug 2026 00:00:00 +0800https://hjjj.top/posts/computer-science/vlm/clip-intro/从双流架构、余弦相似度到 InfoNCE 对比损失，拆解 CLIP 的原理、零样本分类与作为 VLM 底座的应用从 CLIP 到 LLaVA：一文读懂视觉语言模型（VLM）的演进与核心数学公式https://hjjj.top/posts/computer-science/vlm/from-clip-to-llava/Tue, 11 Aug 2026 00:00:00 +0800https://hjjj.top/posts/computer-science/vlm/from-clip-to-llava/从对比学习、生成式指令微调到交叉注意力，拆解 VLM 的架构与三大流派的核心数学原理