Exploring Negatives in Contrastive Learning for Unpaired Image-to-Image Translation

Abstract

Figure 1: Visualization of the learned similarity by the feature extractor. Given an input and output image, we extract the features of these images through a feature extractor. We compute the learned similarities between the feature vectors of [(v, v − 1 ), ..., (v, v − N )] by using exp(v • v − /τ ). Specifically, v is a query element (the highlighted red dot in the output) and [v − 1 , ..., v − N ] are all the candidate patches in the input. Compared with other I2I translation method [28], the feature extractor of our model learns the cross-domain correspondence with a better saliency effect.

Criteria Comparative Learning for Real-scene Image Super-Resolution

“…θ is the parameters of F (•). To obtain different resolution images in real-scene configuration, I hf and I lf are collected by different optical sensors [6], [7], [8], [9], [10] with various resolution settings, which is different from the traditional image super-resolution paradigm [11], [12], [13], [14], [15], [16] that generates I lf using downsampling techniques. Therefore, compared with the traditional image super-resolution task, RealSR suffers a severer pixel displacement due to the difference between the camera settings to obtain I hf and I lf .…”