Learning to Predict Layout-to-image Conditional Convolutions for Semantic Image Synthesis

NeurIPS 2019 Xihui LiuGuojun YinJing ShaoXiaogang WangHongsheng Li

Semantic image synthesis aims at generating photorealistic images from semantic layouts. Previous approaches with conditional generative adversarial networks (GAN) show state-of-the-art performance on this task, which either feed the semantic label maps as inputs to the generator, or use them to modulate the activations in normalization layers via affine transformations... (read more)

PDF Abstract

Results from the Paper


TASK DATASET MODEL METRIC NAME METRIC VALUE GLOBAL RANK RESULT LEADERBOARD
Image-to-Image Translation ADE20K Labels-to-Photos CC-FPSE mIoU 43.7 # 1
Accuracy 82.9% # 1
FID 31.7 # 2
Image-to-Image Translation Cityscapes Labels-to-Photo CC-FPSE Per-pixel Accuracy 82.3% # 1
mIoU 65.5 # 1
FID 54.3 # 2
Image-to-Image Translation COCO-Stuff Labels-to-Photos CC-FPSE mIoU 41.6 # 1
Accuracy 70.7% # 1
FID 19.2 # 1