https://github.com/lukel97/cs4404-project-1
https://github.com/lukel97/cs4404-project-1
cyclegan pytorch tensorflow
Last synced: about 2 months ago
JSON representation
- Host: GitHub
- URL: https://github.com/lukel97/cs4404-project-1
- Owner: lukel97
- Created: 2018-10-02T14:03:54.000Z (almost 8 years ago)
- Default Branch: master
- Last Pushed: 2021-05-21T15:10:13.000Z (about 5 years ago)
- Last Synced: 2025-02-06T19:57:47.116Z (over 1 year ago)
- Topics: cyclegan, pytorch, tensorflow
- Language: Jupyter Notebook
- Size: 180 MB
- Stars: 1
- Watchers: 3
- Forks: 0
- Open Issues: 5
-
Metadata Files:
- Readme: README.md
Awesome Lists containing this project
README
# Guide
Download the datasets first
```bash
./datasets/download.sh
```
## Resources
Original repo: [CycleGAN](https://github.com/junyanz/CycleGAN/)
Comparing implementations in the following frameworks:
* Tensorflow
* PyTorch: [pytorch-CycleGAN-and-pix2pix](https://github.com/junyanz/pytorch-CycleGAN-and-pix2pix)
### PyTorch
Install visdom in order to visualise the training over time:
```
pip install visdom
python -m server.visdom
```
The server will be accessible at https://localhost:8097
#### Set up model
Enter the PyTorch folder using `cd pytorch`.
Download datasets:
``` ./datasets/download_cyclegan_dataset.sh [dataset_name]```
Edit scripts/train_cyclegan.sh with dataset of choice
```bash
set -ex
python3 train.py --dataroot ./datasets/[dataset_name] --name [name] --model cycle_gan --pool_size 50 --no_dropout
```
If your machine is not equipped with an NVIDIA GPU, you can add `--gpu_id -1` to train on the CPU (although this is not recommended and will take a long time).
Checkpoints and samples are saved in the checkpoint folder.
Our CycleGAN setup took ~20 hours to train over 200 epochs on the summer2winter_yosemite dataset.
