https://github.com/xinntao/dni
CVPR19 - Deep Network Interpolation for Continuous Imagery Effect Transition
https://github.com/xinntao/dni
Last synced: about 20 hours ago
JSON representation
CVPR19 - Deep Network Interpolation for Continuous Imagery Effect Transition
- Host: GitHub
- URL: https://github.com/xinntao/dni
- Owner: xinntao
- Created: 2018-11-27T08:51:34.000Z (almost 7 years ago)
- Default Branch: master
- Last Pushed: 2019-04-12T07:54:29.000Z (over 6 years ago)
- Last Synced: 2025-03-29T20:22:51.286Z (7 months ago)
- Homepage: https://xinntao.github.io/projects/DNI
- Size: 451 KB
- Stars: 118
- Watchers: 11
- Forks: 15
- Open Issues: 3
-
Metadata Files:
- Readme: README.md
Awesome Lists containing this project
README
# Deep Network Interpolation [[Paper]](https://arxiv.org/abs/1811.10515) [[Project Page]](https://xinntao.github.io/projects/DNI)
### Deep Network Interpolation for Continuous Imagery Effect Transition
By [Xintao Wang](https://xinntao.github.io/), [Ke Yu](https://yuke93.github.io/), [Chao Dong](https://scholar.google.com.hk/citations?user=OSDCB0UAAAAJ&hl=en), [Xiaoou Tang](https://scholar.google.com/citations?user=qpBtpGsAAAAJ), [Chen Change Loy](http://personal.ie.cuhk.edu.hk/~ccloy/)
#### BibTeX
@Article{wang2018dni,
author={Wang, Xintao and Yu, Ke and Dong, Chao and Tang, Xiaoou and Loy, Chen Change},
title={Deep network interpolation for continuous imagery effect transition},
journal={arXiv preprint arXiv:1811.10515},
year={2018}
}
The following is a YouTube video showing several DNI examples. See [our paper](https://arxiv.org/abs/1811.10515) or the
[project page](https://xinntao.github.io/projects/DNI) for more applications.
### Highlights
1. We propose a simple yet universal approach - deep network interpolation, for smooth and **continuous imagery effect transition without further training**.
2. Different from previous works operating in the feature space, we make an attempt to investigate **the manipulation in the parameter space** of neural networks.
3. Our analyses show that learned filters for several related tasks exhibit continuous changes. We believe that it is worth exploiting the **underlying correlations of learned filters** to further extend the ability and practicality of existing models.
## DNI
DNI is simple and can be implemented with several lines of codes.
```python
alpha = 0.3 # interpolation coefficient
net_A = torch.load('path_to_net_A.pth')
net_B = torch.load('path_to_net_B.pth')
net_interp = OrderedDict()
for k, v_A in net_A.items():
v_B = net_B[k]
net_interp[k] = alpha * v_A + (1 - alpha) * v_B
```