AlphaGAN

This project is an unofficial implementation of AlphaGAN: Generative adversarial networks for natural image matting published at the BMVC 2018. As for now, the result of my experiment is not as good as the paper's.

This is a course project of mine and there may exists some mistakes in this project.

The chinese verison README

Dataset

Adobe Deep Image Matting Dataset

Follow the instruction to contact the author for the dataset

You might need to follow the method mentioned in the Deep Image Matting to generate the trimap using the alpha mat. Here is a algorithm written by someone.

import numpy as np
import cv2 as cv

def generate_trimap(alpha):
    fg = np.array(np.equal(alpha, 255).astype(np.float32))
    # fg = cv.erode(fg, kernel, iterations=np.random.randint(1, 3))
    kernel = cv.getStructuringElement(cv.MORPH_ELLIPSE, (3, 3))
    unknown = np.array(np.not_equal(alpha, 0).astype(np.float32))
    unknown = cv.dilate(unknown, kernel, iterations=np.random.randint(1, 20))
    trimap = fg * 255 + (unknown - fg) * 128
    return trimap.astype(np.uint8)

The Dataset structure in my project

Train
  ├── alpha  # the alpha ground-truth
  ├── bg     # the background image
  ├── fg     # the foreground image
  ├── input  # the real image composed by the fg & bg
  ├── trimap # the trimap

Train & Test

If your project and dataset has the same structure as mine, you could just run the code like below.

python alphaGAN_train.py

You need to specified the dataroot and the folder to save your model

python alphaGAN_train.py --dataroot ${your_dataroot} --save_dir ${your_modelroot}

If you want to visualize the training process, you need to start the visdom first.

python -m visdom.server

The visualization is set to be True by default. You can disable the visualization by

python alphaGAN_train.py --visual 0

For more details

python alphaGAN_train.py --help

Acknowledgments

My code is inspired by:

• pytorch-CycleGAN-and-pix2pix

• pytorch-book chapter7 generate anime head portrait with GAN

• pytorch-deeplab-xception

Test Result

Although I train this network with the fixed size input(320*320), it can still process different size of images.

The testing result after 15 epoch

After 35 epoch

Ground-truth

Trimap