Weakly- and Self-Supervised Learning for Content-Aware Deep Image Retargeting

Abstract

This paper proposes a weakly-and self-supervised deep convolutional neural network (WSSDCNN) for contentaware image retargeting. Our network takes a source image and a target aspect ratio, and then directly outputs a retargeted image. Retargeting is performed through a shift map, which is a pixel-wise mapping from the source to the target grid. Our method implicitly learns an attention map, which leads to a content-aware shift map for image retargeting. As a result, discriminative parts in an image are preserved, while background regions are adjusted seamlessly. In the training phase, pairs of an image and its image-level annotation are used to compute content and structure losses. We demonstrate the effectiveness of our proposed method for a retargeting application with insightful analyses. AbstractThis paper proposes weakly-and self-supervised deep convolutional neural networks (WSSDCNN) for contentsaware image retargeting. Our network takes a source image and a target aspect ratio, and then directly produces a retargeted image. Retargeting is performed through a shift map which is a pixel-wise mapping from source to target grid. Our method implicitly learns an attention map, which leads to a context-aware shift map for warping As a result, discriminative parts in an image are preserved, while background regions are adjusted seamlessly. In training phase, pairs of an image and its image level annotation used for computation of content and structure losses. We demonstrate effectiveness of the proposed method for the retargeting application with insightful analyses.

Cycle-IR: Deep Cyclic Image Retargeting

“…11 shows more visual comparisons on RetargetMe dataset. It intuitively shows that significant distortions often appear in the retargeted results of Cho et al [25] method. In contrast, benefiting from the constraint of cyclic perception consistency, our Cycle-IR avoids this serious problem.…”

“…For example, in [25] method, the men's hair in "glass" image is reduced too much, while our approach preserves hair better. Besides, the men's forehead has some distortion in [25] method. For "eagle" image, our approach preserves better for hawk's wings than [25] method.…”

“…Besides, the men's forehead has some distortion in [25] method. For "eagle" image, our approach preserves better for hawk's wings than [25] method.…”

“…Therefore, using deep learning techniques to solve image retargeting problem becomes an extremely challenging problem. Cho et al [25] make an attempt to use deep learning technique to addressing image retargeting problem. They propose a weakly and self-supervised deep network to learn a shift map for manipulating the aspect ratio of input images.…”

A2-RL: Aesthetics Aware Reinforcement Learning for Image Cropping

“…There are also change-based methods [34], which compares original images with cropped images so as to throw away distracting regions and retain high quality ones. Image retargeting techniques [6,3] adjust the aspect ratio of an image to fit the target aspect ratio, while not discarding important content in an image, which are relevant to our task.…”