0000−0001−8949−9597] , Aiwen Jiang 1[0000−0002−5979−7590] , Juncheng Li 2[0000−0001−7314−6754] , and Abstract. Single image dehazing is a challenging ill-posed restoration problem. Various prior-based and learning-based methods have been proposed. Most of them follow a classic atmospheric scattering model which is an elegant simplified physical model based on the assumption of singlescattering and homogeneous atmospheric medium. The formulation of haze in realistic environment is more complicated. In this paper, we propose to take its essential mechanism as "black box", and focus on learning an input-adaptive trainable end-to-end dehazing model. An U-Net like encoder-decoder deep network via progressive feature fusions has been proposed to directly learn highly nonlinear transformation function from observed hazy image to haze-free ground-truth. The proposed network is evaluated on two public image dehazing benchmarks. The experiments demonstrate that it can achieve superior performance when compared with popular state-of-the-art methods. With efficient GPU memory usage, it can satisfactorily recover ultra high definition hazed image up to 4K resolution, which is unaffordable by many deep learning based dehazing algorithms.

Object detection has so far achieved great success. However, almost all of current state-ofthe-art methods focus on images with normal illumination, while object detection under low-illumination is often ignored. In this paper, we have extensively investigated several important issues related to the challenge low-illumination detection task, such as the importance of illumination on detection, the applicabilities of illumination enhancement on low-illumination object detection task, and the influences of illumination balanced dataset and model's parameters initialization, etc. We further have proposed a Night Vision Detector (NVD) with specifically designed feature pyramid network and context fusion network for object detection under low-illuminance. Through conducting comprehensive experiments on a public real low-illuminance scene dataset ExDARK and a selected normal-illumination counterpart COCO*, we on one hand have reached some valuable conclusions for reference, on the other hand, have found specific solutions for low-illumination object detection. Our strategy improves detection performance by 0.5%~2.8% higher than basic model on all standard COCO evaluation criterions. Our work can be taken as effective baseline and shed light to future studies on low-illumination detection.

No abstract

0000−0001−8949−9597] , Aiwen Jiang 1[0000−0002−5979−7590] , Juncheng Li 2[0000−0001−7314−6754]Abstract. Recent works on single-image super-resolution are concentrated on improving performance through enhancing spatial encoding between convolutional layers. In this paper, we focus on modeling the correlations between channels of convolutional features. We present an effective deep residual network based on squeeze-and-excitation blocks (SEBlock) to reconstruct high-resolution (HR) image from low-resolution (LR) image. SEBlock is used to adaptively recalibrate channel-wise feature mappings. Further, short connections between each SEBlock are used to remedy information loss. Extensive experiments show that our model can achieve the state-of-the-art performance and get finer texture details.

No abstract