A Systematic Performance Analysis of Deep Perceptual Loss Networks: Breaking Transfer Learning Conventions

Deep perceptual loss is a type of loss function in computer vision that aims to mimic human perception by using the deep features extracted from neural networks. In recent years, the method has been applied to great effect on a host of interesting computer vision tasks, especially for tasks with image or image-like outputs, such as image synthesis, segmentation, depth prediction, and more. Many applications of the method use pretrained networks, often convolutional networks, for loss calculation. Despite the increased interest and broader use, more effort is needed toward exploring which networks to use for calculating deep perceptual loss and from which layers to extract the features. This work aims to rectify this by systematically evaluating a host of commonly used and readily available, pretrained networks for a number of different feature extraction points on four existing use cases of deep perceptual loss. The use cases of perceptual similarity, super-resolution, image segmentation, and dimensionality reduction, are evaluated through benchmarks. The benchmarks are implementations of previous works where the selected networks and extraction points are evaluated. The performance on the benchmarks, and attributes of the networks and extraction points are then used as a basis for an in-depth analysis. This analysis uncovers insight regarding which architectures provide superior performance for deep perceptual loss and how to choose an appropriate extraction point for a particular task and dataset. Furthermore, the work discusses the implications of the results for deep perceptual loss and the broader field of transfer learning. The results show that deep perceptual loss deviates from two commonly held conventions in transfer learning, which suggests that those conventions are in need of deeper analysis.