The ability to estimate the perceptual error between images is an important problem in computer vision with many applications. Although it has been studied extensively, however, no method currently exists that can robustly predict visual differences like humans. Some previous approaches used hand-coded models, but they fail to model the complexity of the human visual system. Others used machine learning to train models on human-labeled datasets, but creating large, high-quality datasets is difficult because people are unable to assign consistent error labels to distorted images. In this paper, we present PieAPP, a metric that predicts perceptual error of a distorted image with respect to a reference in a manner consistent with human observers.

Since it is much easier for people to compare two given images and identify the one more similar to a reference than to assign quality scores to each, we propose a new, large-scale dataset labeled with the probability that humans will prefer one image over another. We then train a deep-learning model using a novel, pairwise-learning framework to predict the preference of one distorted image over the other. Our key observation is that our trained network can then be used separately with only one distorted image and a reference to predict its perceptual error, without ever being trained on explicit human perceptual-error labels. The perceptual error estimated by PieAPP is well-correlated with human opinion. Furthermore, it significantly outperforms existing algorithms, while also generalizing to new kinds of distortions, unlike previous learning-based methods.

Technical details about PieAPPv0.1:

Try out PieAPPv0.1:

Try out the PieAPP dataset:

Method	KRCC		PLCC	SRCC
MAPE	0.233	0.278	0.170	0.290
MRSE	0.185	0.211	0.135	0.219
NQM	0.216	0.234	0.508	0.246
IFC	0.165	0.178	0.230	0.198
VIF	0.179	0.192	0.250	0.212
MAD	0.270	0.301	0.231	0.304
RFSIM	0.217	0.246	0.368	0.247
GSM	0.324	0.376	0.356	0.378
SR-SIM	0.296	0.356	0.352	0.358
MDSI	0.280	0.336	0.349	0.350
MAE	0.252	0.289	0.302	0.302
RMSE	0.289	0.339	0.324	0.351
SSIM	0.272	0.323	0.245	0.316
MS-SSIM	0.275	0.325	0.051	0.321
GMSD	0.250	0.291	0.242	0.297
PSNR-HMA	0.245	0.274	0.310	0.281
FSIMc	0.322	0.377	0.481	0.378
SFF	0.258	0.295	0.025	0.305
SCQI	0.303	0.364	0.267	0.360
DOG-SSIMc	0.263	0.320	0.417	0.464
Lukin et al.	0.290	0.396	0.496	0.386
Kim et al. (CVPR 2017)	0.211	0.240	0.172	0.252
Bosse et al. NR (TIP 2018)	0.269	0.353	0.439	0.352
Bosse et al. FR (TIP 2018)	0.414	0.503	0.568	0.537
LPIPS: VGG-lin (CVPR 2018)	0.503	0.625	0.654	0.641
LPIPS: Alex-lin (CVPR 2018)	0.485	0.601	0.644	0.625
LPIPS: Squeeze-lin (CVPR 2018)	0.500	0.624	0.678	0.647
PieAPPv0.1 (CVPR 2018)	0.668	0.815	0.842	0.831