Rain removal using cycle-consistency adversarial network
Raindrops in videos and images can hamper the visibility of objects in a scene, leading to a loss of video quality. In this project, we address the problem of rain removal in images by using an unsupervised learning approach relying on a new framework of cycle-consistent generative adversarial netwo...
Saved in:
Main Author: | |
---|---|
Other Authors: | |
Format: | Final Year Project |
Language: | English |
Published: |
2019
|
Subjects: | |
Online Access: | http://hdl.handle.net/10356/76949 |
Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
Institution: | Nanyang Technological University |
Language: | English |
Summary: | Raindrops in videos and images can hamper the visibility of objects in a scene, leading to a loss of video quality. In this project, we address the problem of rain removal in images by using an unsupervised learning approach relying on a new framework of cycle-consistent generative adversarial networks. Unlike usual image domain transfer problem, the proposed solution solves the problem by having two asymmetric functions: a forward function that removes the rain from a rain degraded image and a backward function that adds rain into a rain-free clean image. The main idea is to have two coupled generative adversarial network that implements these two functions: one that would remove rain from a rain degraded image and a second network that would add rain into a rain-free clean image. Our experiments show the effectiveness of our approach and how it performs against other previous works. |
---|