Learning from Simulated and Unsupervised Images through Adversarial Training

CVPR 2017 Ashish ShrivastavaTomas PfisterOncel TuzelJosh SusskindWenda WangRuss Webb

With recent progress in graphics, it has become more tractable to train models on synthetic images, potentially avoiding the need for expensive annotations. However, learning from synthetic images may not achieve the desired performance due to a gap between synthetic and real image distributions... (read more)

PDF Abstract

Evaluation results from the paper


Task Dataset Model Metric name Metric value Global rank Compare
Image-to-Image Translation Cityscapes Labels-to-Photo SimGAN Class IOU 0.04 # 4
Image-to-Image Translation Cityscapes Labels-to-Photo SimGAN Per-class Accuracy 10% # 3
Image-to-Image Translation Cityscapes Labels-to-Photo SimGAN Per-pixel Accuracy 20% # 8
Image-to-Image Translation Cityscapes Photo-to-Labels SimGAN Per-pixel Accuracy 47% # 3
Image-to-Image Translation Cityscapes Photo-to-Labels SimGAN Per-class Accuracy 11% # 4
Image-to-Image Translation Cityscapes Photo-to-Labels SimGAN Class IOU 0.07 # 4