README.md

About

This repo is the image manipulation demo of "Self-Supervised Scene De-occlusion".

• Below is the demo.

For further information, please contact Xiaohang Zhan.

Usage

1. Clone the repo, install dependencies.

   git clone https://github.com/XiaohangZhan/deocclusion-demo
   pip install PyQt5, opencv-python
   

2. Run the demo.

   python main.py
   

3. Interactions.

   • Click Open to open an image from decomposition/image_*.png.
   • Click De-occlusion to load decomposed components.
   • Use mouse to drag objects.
   • Mouse right click to change ordering or save the object.
   • Click Show Objects to show each object in order.
   • Click Reset to reset to the original status.
   • Click Insert and open an image from materials to insert it.
   • Push Up or Down arrow button to zoom out or zoom in the object.
   • Push Left or Right arrow button to rotate the object.
   • Click Save As to save the re-composed image.

4. Try new images.

• First of all, you should launch the jupyter notebooks here, e.g., demo_cocoa.ipynb.

• Second, run the notebook up to the last cell. In the last cell, change the False under # save to True. In this way, the completed objects as well as the background are saved following the topological order under outputs/decomposition/.

• Third, copy the image as well as the folder containing the decomposed components under decomposition in this repo. Then enjoy yourself to re-compose the image.

Notice

There are still some bugs. Since I have no time to fix them, you are welcome to raise pull request to fix them. The bugs are below:

• When you click Open or Insert to browse a folder, but choose cancel, the program crash.