Transformer based Pluralistic Image Completion with Reduced Information Loss

They quantize 256^3 RGB values to a small number (such as 512) of quantized color values. The indices of quantized pixels are used as tokens for the inputs and prediction targets of the transformer.
In addition, pluralistic images can be produced when the content of masked regions is predicted and sampled in an autoregressive manner.

posted @   Trkly  阅读(3)  评论(0编辑  收藏  举报
相关博文:
阅读排行:
· 无需6万激活码!GitHub神秘组织3小时极速复刻Manus,手把手教你使用OpenManus搭建本
· C#/.NET/.NET Core优秀项目和框架2025年2月简报
· 葡萄城 AI 搜索升级:DeepSeek 加持,客户体验更智能
· 什么是nginx的强缓存和协商缓存
· 一文读懂知识蒸馏
点击右上角即可分享
微信分享提示