https://github.com/wtupc96/copy-pasting-gan

Object Discovery with a Copy-Pasting GAN
https://github.com/wtupc96/copy-pasting-gan

copy-paste gan image-processing mask tensorflow

Last synced: about 1 year ago
JSON representation

Object Discovery with a Copy-Pasting GAN

• Host: GitHub
• URL: https://github.com/wtupc96/copy-pasting-gan
• Owner: wtupc96
• License: mit
• Created: 2019-07-25T02:34:43.000Z (almost 7 years ago)
• Default Branch: master
• Last Pushed: 2019-12-28T07:10:39.000Z (over 6 years ago)
• Last Synced: 2025-02-25T09:24:00.755Z (over 1 year ago)
• Topics: copy-paste, gan, image-processing, mask, tensorflow
• Language: Python
• Size: 232 KB
• Stars: 8
• Watchers: 0
• Forks: 1
• Open Issues: 2
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

README

          
# Copy-Pasting GAN

A TensorFlow implementation of paper **Object Discovery with a Copy-Pasting GAN**.

## Requirements

- Python 3.7

- TensorFlow 1.13.1

## Run

1. Prepare dataset

   1. Prepare foreground and background images(`jpg` format only);

   2. Put foreground images into `data/plane_sky/plane`;

   3. Put background images into `data/plane_sky/sky`;

   4. Resize foreground and background images to $240\times240$;(You can use `img_utils.py` in `utils` folder);

   - You can change image format in `train.py` and `utils/img_utils.py`;

   - You can change image folder in `cfgs.py`;

2. Train with `python main.py`.

3. You can find checkpoints and saved images in `logs`. :)

## Results

| ![foreground](imgs/foreground_img_0.jpg) | ![background](imgs/background_img_0.jpg) | ![shuffled_foreground](imgs/shuffled_foreground_img_0.jpg) |

| :----------------------------------------------------------: | :----------------------------------------------------------: | :----------------------------------------------------------: |

|                          Foreground                          |                          Background                          |                  Random selected foreground                  |

| ![foreground](imgs/random_mask_img_0.jpg) | ![background](imgs/grounded_fake_img_0.jpg) | ![shuffled_foreground](imgs/d_mask_grounded_fake_img_0.jpg) |

|                    Random generated Mask                     |             Grounded Fake
(with random mask)              |         Predicted mask from D
(for Grounded Fake)         |

| ![foreground](imgs/g_mask_foreground_img_0.jpg) | ![background](imgs/anti_shortcut_img_0.jpg) | ![shuffled_foreground](imgs/d_mask_anti_shortcut_img_0.jpg) |

|          Predicted mask from G
(for Foreground)           |       Anti-shortcut
(foreground+random foreground)        |         Predicted mask from D
(for Anti-shortcut)         |

| ![foreground](imgs/d_mask_foreground_img_0.jpg) | ![background](imgs/composited_img_0.jpg) | ![shuffled_foreground](imgs/d_mask_composite_img_0.jpg) |

|          Predicted mask from D
(for Foreground)           |       **Composited image**
(foreground+background)        |       Predicted mask from D
(for Composited image)        |