Single Image Super-Resolution (SISR) based on deep learning methods has been widely studied for applications on remote sensing images. With limited remote sensing images, most of the existing SISR methods simply adopt the regular data augmentation approaches (such as flip) in natural images to improve model performance. Considering the fact that remote sensing images are all taken from a bird’s-eye view and objects appear in multiple directions, we first introduce rotation augmentation method in remote sensing images to promote diversity of samples dramatically, as rotation does not cause semantic problems like people standing upside down in natural images. However, image rotation at various angles implemented by interpolation will cause the inconsistent pixel distribution problem for the pixel level task. Thus, we propose Transformation Consistency Loss Function (TCLF) to narrow the gap between the augmented and original distribution, while expanding the feature space with rotation augmentation method. Extensive experiments are performed on UC-Merced Land-use dataset of 21 remote sensing scenes, and the results as well as ablation studies demonstrate our proposed method outperforms mainstream methods.

High-quality remote sensing images are difficult to obtain due to limited conditions and high cost for data acquisition. With the development of machine vision and deep learning, some image generation methods (e.g., GANs) are introduced into this field, but it’s still hard to generate images with good texture details and structural dependencies. We establish Skip Attention Mechanism to deal with this problem, which learns dependencies between local points on low-resolution feature maps, and then upsample the attention map and combine it with high-resolution feature maps. With this method, long-range dependencies learned from low-resolution are used for generating remote sensing images with more structural details. We name this method as Skip Attention GAN, which is the first method applying cross-scale attention mechanism for unsupervised remote sensing image generation. Experiments show that our method outperforms previous methods under several metrics. Visual and ablation results of attention layers show that Skip Attention has learned long-distance structural dependencies between similar targets.

We propose a simple yet effective method to reduce the redundancy of DenseNet by substantially decreasing the number of stacked modules by replacing the original bottleneck by our SMG module, which is augmented by local residual. Furthermore, SMG module is equipped with an efficient two-stage pipeline, which aims to DenseNet-like architectures that need to integrate all previous outputs, i.e., squeezing the incoming informative but redundant features gradually by hierarchical convolutions as a hourglass shape and then exciting it by multi-kernel depthwise convolutions, the output of which would be compact and hold more informative multi-scale features. We further develop a forget and an update gate by introducing the popular attention modules to implement the effective fusion instead of a simple addition between reused and new features. Due to the Hybrid Connectivity (nested combination of global dense and local residual) and Gated mechanisms, we called our network as the HCGNet. Experimental results on CIFAR and ImageNet datasets show that HCGNet is more prominently efficient than DenseNet, and can also significantly outperform state-of-the-art networks with less complexity. Moreover, HCGNet also shows the remarkable interpretability and robustness by network dissection and adversarial defense, respectively. On MS-COCO, HCGNet can consistently learn better features than popular backbones.

Recently, multi-resolution networks (such as Hourglass, CPN, HRNet, etc.) have achieved significant performance on pose estimation by combining feature maps of various resolutions. In this paper, we propose a Resolution-wise Attention Module (RAM) and Gradual Pyramid Refinement (GPR), to learn enhanced resolution-wise feature maps for precise pose estimation. Specifically, RAM learns a group of weights to represent the different importance of feature maps across resolutions, and the GPR gradually merges every two feature maps from low to high resolutions to regress final human keypoint heatmaps. With the enhanced resolution-wise features learnt by CNN, we obtain more accurate human keypoint locations. The efficacies of our proposed methods are demonstrated on MS-COCO dataset, achieving state-of-the-art performance with average precision of 77.7 on COCO va12017 set and 77.0 on test-dev2017 set without using extra human keypoint training dataset.