Cycle-IR: Deep Cyclic Image Retargeting

Abstract

Supervised deep learning techniques have achieved great success in various fields due to getting rid of the limitation of handcrafted representations. However, most previous image retargeting algorithms still employ fixed design principles such as using gradient map or handcrafted features to compute saliency map, which inevitably restricts its generality. Deep learning techniques may help to address this issue, but the challenging problem is that we need to build a large-scale image retargeting dataset for the training of deep retargeting models. However, building such a dataset requires huge human efforts.In this paper, we propose a novel deep cyclic image retargeting approach, called Cycle-IR, to firstly implement image retargeting with a single deep model, without relying on any explicit user annotations. Our idea is built on the reverse mapping from the retargeted images to the given input images. If the retargeted image has serious distortion or excessive loss of important visual information, the reverse mapping is unlikely to restore the input image well. We constrain this forward-reverse consistency by introducing a cyclic perception coherence loss. In addition, we propose a simple yet effective image retargeting network (IRNet) to implement the image retargeting process. Our IRNet contains a spatial and channel attention layer, which is able to discriminate visually important regions of input images effectively, especially in cluttered images. Given arbitrary sizes of input images and desired aspect ratios, our Cycle-IR can produce visually pleasing target images directly. Extensive experiments on the standard RetargetMe dataset show the superiority of our Cycle-IR. In addition, our Cycle-IR outperforms the Multiop method and obtains the best result in the user study. Code is available at https://github.com/mintanwei/Cycle-IR.

Self-Play Reinforcement Learning for Fast Image Retargeting

“…Table 1 shows the average vote rate. Our method obtained a higher vote rate than scaling, GAIC [33], seam carving [21], WSSDCNN [3], and Cycle-IR [26]; this shows that our method is subjectively superior to these methods. Furthermore, there was almost no difference in the average vote rate between the MULTIOP [22] and our method, showing that the performance of our method was subjectively similar to that of MULTIOP.…”

“…Cho et al [3] proposed a weakly-and self-supervised learning model (WSSDCNN) that learns the shift map of each pixel in the input and output images. Tan et al [26] proposed an unsupervised learning model, Cycle-IR, which learned the forward and reverse mapping of input and output images. Lee et al [14] used object detection and object tracking with a deep neural network to enable consistent video retargeting.…”

Fast Hybrid Image Retargeting

“…AAD denotes the warponly method described in [1] but employing importance maps produced by our method, as described in Section 3.1 (instead of using the hand-crafted maps utilised in [1]). CycleIR is a recent DNN-based approach [6]. Among the retargeting approaches compared in Fig.…”

“…Our implementation has been tested both on desktop and on mobile devices, and a user study based on the Re-targetMe benchmark [4] shows that our method outperforms The authors wish to thank Mete Ozay for his advice and support. recent approaches, while executing in a fraction of the time [5,6,7,8,9,10].…”

PAMSGAN: Pyramid Attention Mechanism-Oriented Symmetry Generative Adversarial Network for Motion Image Deblurring

“…are encouraged by training mapping G and F simultaneously with adding cyclic consistency loss [27]. This loss is combined with the adversarial loss and the content loss to train the cyclic consistent feature transfer algorithm.…”