基于上下文的编码器：图像修复的特征学习 Wepresentanunsupervisedvisualfeaturelearningalgo-rithmdrivenbycontext-basedpixelprediction.Byanalogywithauto-encoders,weproposeContextEncoders–acon-volutionalneuralnetworktrainedtogeneratethecontentsofanarbitraryimageregionconditionedonits