A neural network based on Conditional GAN which can translated image to colored model.
I implemented a system to show the results by React and Three.js. (The codes are not included in this file)
git clone https://github.com/ZijingPeng/3d-cgan.git
I downloaded the datasets from im2avatar and 3D-R2N2 , and made some revise (see details in prepare_dataset.py).
You can just download the dataset I prepared:
- ShapeNetCore_im2avatar: https://drive.google.com/open?id=1swd-K0iBFK08LfOWo3MF_63YGZ22-xwc
Download pretrained checkpoints here.
Download checkpoints and copy to ckp folder, download datasets to ShapeNetCore_im2avatar folder.
Then run the test.py to generate models or run train.py to train.
When you are testing images downloaded online, you had better use cv2.dilate or similar function to preprocess the images, in order to exclude light influences. (see comments in test.py )
MIT License