Image-to-image translation based on generative models
Image-to-image translation tasks have become a widely studied topic in computer vision. Image-to-image translation aims at finding a model that is fed with the input image and generating desired output image correspondingly. Previous studies that are based on deep neural networks were mostly built u...
Saved in:
| Main Author: | |
|---|---|
| Other Authors: | |
| Format: | Thesis-Master by Coursework |
| Language: | English |
| Published: |
Nanyang Technological University
2022
|
| Subjects: | |
| Online Access: | https://hdl.handle.net/10356/154672 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Institution: | Nanyang Technological University |
| Language: | English |
| Summary: | Image-to-image translation tasks have become a widely studied topic in computer vision. Image-to-image translation aims at finding a model that is fed with the input image and generating desired output image correspondingly. Previous studies that are based on deep neural networks were mostly built upon
encoder-decoder architecture, where a direct mapping from input to target output is learned, without exploring the distribution of images. In this thesis, generative models are used to capture the distribution of images, and the potentials of generative models on the image-to-image translation tasks are explored. Specifically, an improved CycleGAN is proposed to conduct the style transfer task and a DDPM-based conditional generative model is used for image colorization. Empirical results show that the generative models can achieve competitive results in image-to-image translation tasks. |
|---|