[图像处理]高底层任务结合论文归档

高层任务如Recognition, Detection, Tracking, Segmentation等，底层任务如Derainning, Denoising, Deblurring等等。高底层任务长期以来都是相对分离的两个领域，分开来进行研究。近期越来越多的研究提出，以PSNR, SSMI等指标评价的底层任务并不一定能很好的为高层任务服务。那么，什么样的去雨、去雾、去噪结果才是高层任务真正想要的呢，目前为止还没有统一的说法。另外，目前的高底层任务训练大多数只能在仿真数据集上训练，拥有退化图像-干净图像-高层任务标注的数据是非常稀缺的，因此也成为了这个topic的难点之一。

以Deep Learning文章为主，将近年来的paper整理如下：

Low level for high level:

[1] 2019_TIP Connecting Image Denoising and High-Level Vision Tasks via Deep Learning(Liu Ding)

[2] 2017_ICCV_AOD-Net All-in-One Dehazing Network

[3] 2018_IJCAI When Image Denoising Meets High-Level Vision Tasks A Deep Learning Approach

[4] 2019_arKiv_Segmentation-Aware Image Denoising without Knowing True Segmentation

[5] 2019_arKiv_Bridging the Gap Between Computational Photography and Visual Recognition.pdf

High level for low level:

[6] 2018_TIP_Class-Aware Fully-Convolutional Gaussian and poisson Denoising

[7] 2018_ECCV_A Segmentation-aware Deep Fusion Network for Compressed Sensing MRI

[8] 2018_CVPR Classification-Driven Dynamic Image Enhancement

[9] 2017_ICCV_AOD-Net All-in-One Dehazing Network