Residual and Dense UNet for Under-Display Camera Restoration
Abstract
With the rapid development of electronic products, the increasing demand for full-screen devices has become a new trend, which facilitates the investigation of Under-Display Cameras (UDC). UDC can not only bring larger display-to-body ratio but also improve the interactive experience. However, when imaging sensor is mounted behind a display, existing screen materials will cause severe image degradation due to lower light transmission rate and diffraction effects. In order to promote the research in this field, RLQ-TOD 2020 held the Image Restoration Challenge for Under-Display Camera. The challenge was composed of two tracks – 4k Transparent OLED (T-OLED) and phone Pentile OLED (P-OLED) track. In this paper, we propose a UNet-like structure with two various basic building blocks to tackle this problem. We discover that T-OLED and P-OLED have different preferences with the model structure and the input patch size during training. With the proposed model, our team won the third place in the challenge on both T-OLED and P-OLED tracks.
Cite
Text
Yang et al. "Residual and Dense UNet for Under-Display Camera Restoration." European Conference on Computer Vision Workshops, 2020. doi:10.1007/978-3-030-68238-5_30Markdown
[Yang et al. "Residual and Dense UNet for Under-Display Camera Restoration." European Conference on Computer Vision Workshops, 2020.](https://mlanthology.org/eccvw/2020/yang2020eccvw-residual/) doi:10.1007/978-3-030-68238-5_30BibTeX
@inproceedings{yang2020eccvw-residual,
title = {{Residual and Dense UNet for Under-Display Camera Restoration}},
author = {Yang, Qirui and Liu, Yihao and Tang, Jigang and Ku, Tao},
booktitle = {European Conference on Computer Vision Workshops},
year = {2020},
pages = {398-408},
doi = {10.1007/978-3-030-68238-5_30},
url = {https://mlanthology.org/eccvw/2020/yang2020eccvw-residual/}
}