Pretraining Image Encoders without Reconstruction via Feature Prediction Loss

This work investigates three different loss functions for autoencoder-based pretraining of image encoders: The commonly used reconstruction loss, the more recently introduced perceptual similarity loss, and a feature prediction loss proposed here; the latter turning out to be the most efficient choice. Former work shows that predictions based on embeddings generated by image autoencoders can be improved by training with perceptual loss. So far the autoencoders trained with perceptual loss networks implemented an explicit comparison of the original and reconstructed images using the loss network. However, given such a loss network we show that there is no need for the timeconsuming task of decoding the entire image. Instead, we propose to decode the features of the loss network, hence the name "feature prediction loss". To evaluate this method we compare six different procedures for training image encoders based on pixel-wise, perceptual similarity, and feature prediction loss. The embedding-based prediction results show that encoders trained with feature prediction loss is as good or better than those trained with the other two losses. Additionally, the encoder is significantly faster to train using feature prediction loss in comparison to the other losses. The method implementation used in this work is available online: this https URL