📄 中文摘要
一种新的深度图像修复方法能够有效填补照片中的缺失部分，使得图像看起来完好无损。该方法利用深度模型分析图像的其余部分，理解场景内容，并用合适的内容填补空白。其核心在于平衡保留已有内容与生成新部分的真实感，使得人脸或物体能够自然融合。经过在面孔和街道号码图像上的测试，该系统能够在大部分图像缺失或大块区域缺失的情况下恢复细节，甚至能够准确预测面部特征，达到无缝修复的效果。该系统将照片映射为小代码，进而进行修复。
基于感知和上下文损失的语义图像修复
出处: Semantic Image Inpainting with Perceptual and Contextual Losses
发布: 2026年2月16日
📄 English Summary
Semantic Image Inpainting with Perceptual and Contextual Losses
A new deep image inpainting method effectively fills in missing parts of photos, making them appear whole again. This approach uses a deep model to analyze the remaining parts of the image, understand the scene context, and fill gaps with appropriate content. The key lies in balancing the retention of existing content with the realism of the newly generated parts, allowing faces or objects to blend naturally. Tested on images of faces and street numbers, the system can recover details even when most of the image is missing or large blocks are absent, and it can accurately predict facial features, achieving seamless restoration. The system maps a photo to a small code for the inpainting process.