This is the project about our JSHDR (CVPR2021). Our paper is here [[https://openaccess.thecvf.com/content/CVPR2021/papers/Fu_A_Multi-Task_Network_for_Joint_Specular_Highlight_Detection_and_Removal_CVPR_2021_paper.pdf][A Multi-Task Network for Joint Specular Highlight Detection and Removal]].
@InProceedings{fu-2021-multi-task, author = {Fu, Gang and Zhang, Qing and Zhu, Lei and Li, Ping and Xiao, Chunxia}, title = {A Multi-Task Network for Joint Specular Highlight Detection and Removal}, booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)}, year = {2021}, pages = {7752-7761}, month = {June}, tags = {CVPR}, }
Overview [[./images/highlight_removal.png]]
Figure 1: Visual comparison of highlight detection and removal on an example image from our dataset. We compare our method with the state-of-the-art removal method Shi /et al/. [7], Guo /et al/. [6], and Yang /et al/ [3], and with the state-of-the-art detection methods including Zhang /et al/. [2], Li /et al/ [1], and Fu /et al/ [8]. Please zoom in to view fine details.
Specular highlight detection and removal are fundamental and challenging tasks. Although recent methods have achieved promising results on the two tasks by training on synthetic training data in a supervised manner, they are typically solely designed for highlight detection or removal, and their performance usually deteriorates significantly on real-world images. In this paper, we present a novel network that aims to detect and remove highlights from natural images. To remove the domain gap between synthetic training samples and real test images, and support the investigation of learning-based approaches, we first introduce a dataset with about 16K real images, each of which has the corresponding ground truths of highlight detection and removal. Using the presented dataset, we develop a multi-task network for joint highlight detection and removal, based on a new specular highlight image formation model. Experiments on the benchmark datasets and our new dataset show that our approach clearly outperforms state-of-the-art methods for both highlight detection and removal.
Specular highlight image quadruples (SHIQ)
To enable effective training and comprehensive evaluation for highlight detection and removal, we in this work introduce a large-scale real dataset for highlight detection and removal. It covers a wide range of scenes, subjects, and lighting conditions. Each image in the dataset has the corresponding highlight detection, removal, and highlight intensity images. Several examples in our dataset are shown in Figure 2.
[[./images/data_teaser.png]]
Figure 2: An illustration of several highlight (1st row), highlight-free (2nd row), highlight intensity (3rd row) and highlight mask (4th row) image quadruples in our dataset.
Some notes about the data generation as follows:
@InProceedings{fu-2021-multi-task, author = {Fu, Gang and Zhang, Qing and Zhu, Lei and Li, Ping and Xiao, Chunxia}, title = {A Multi-Task Network for Joint Specular Highlight Detection and Removal}, booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)}, year = {2021}, pages = {7752-7761}, month = {June}, tags = {CVPR}, }
Network ** Requirements
References