Unverified50% confidenceFactExact time
ViT-S版本嵌入维度为384，为每个16×16图像块输出一个384维特征向量
1
Sources
50%
Confidence
Long-term
Relevance
7/12/2026
First Seen
Sources
自监督视觉模型PCA可视化避坑指南：颜色为何毫无意义
redditr/learnmachinelearning7/11/2026
Related Claims
UnverifiedCLIP ViT-L参数量约3亿，嵌入维度为76871% similarUnverified千问3.8 27B是27.8B的稠密模型，支持图像和视频原生多模态输入，原生262K上下文67% similarUnverifiedQwen3.6 27B支持262K上下文长度，34B 80层版本保持相同的上下文长度不变65% similarUnverified剪辑用途优先选32英寸4K（3840×2160）而非27英寸4K，27英寸4K像素密度约163PPI在100%缩放时时间线文字过小需缩放，32英寸约138PPI可100%无缩放显示更多轨道65% similarUnverifiedViT-L模型约3亿参数，checkpoint达1213MB，CPU上处理每张图像需4.5秒65% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/490400API
curl https://kongchang.com/api/v1/knowledge/claims/490400MCP
get_claim(id=490400)