Semantic Image Synthesis with Spatially-Adaptive Normalization

18 Mar 2019Taesung ParkMing-Yu LiuTing-Chun WangJun-Yan Zhu

We propose spatially-adaptive normalization, a simple but effective layer for synthesizing photorealistic images given an input semantic layout. Previous methods directly feed the semantic layout as input to the deep network, which is then processed through stacks of convolution, normalization, and nonlinearity layers... (read more)

PDF Abstract
Task Dataset Model Metric name Metric value Global rank Compare
Image-to-Image Translation ADE20K Labels-to-Photos SPADE mIoU 38.5 # 1
Image-to-Image Translation ADE20K Labels-to-Photos SPADE Accuracy 79.9% # 1
Image-to-Image Translation ADE20K Labels-to-Photos SPADE FID 33.9 # 1
Image-to-Image Translation ADE20K-Outdoor Labels-to-Photos SPADE mIoU 30.8 # 1
Image-to-Image Translation ADE20K-Outdoor Labels-to-Photos SPADE Accuracy 82.9% # 1
Image-to-Image Translation ADE20K-Outdoor Labels-to-Photos SPADE FID 63.3 # 1
Image-to-Image Translation Cityscapes Labels-to-Photo SPADE Per-pixel Accuracy 81.9% # 1
Image-to-Image Translation Cityscapes Labels-to-Photo SPADE mIoU 62.3 # 1
Image-to-Image Translation Cityscapes Labels-to-Photo SPADE FID 71.8 # 2
Image-to-Image Translation COCO-Stuff Labels-to-Photos SPADE mIoU 37.4 # 1
Image-to-Image Translation COCO-Stuff Labels-to-Photos SPADE Accuracy 67.9% # 1
Image-to-Image Translation COCO-Stuff Labels-to-Photos SPADE FID 22.6 # 1