FCNet

FCNet encompasses three colorization approaches, i.e., single- or multi-reference image-guided colorization in (a), sample-guided colorization in (b), and automatic colorization in (c). In (a), the first column is the grayscale input, and the subsequent five columns show the reference and the corresponding single-reference colorization results. Then, the last column is a multi-reference colorization result, taking colors from different facial components of the references. In (b), the results are generated according to the sampled single or multiple-color representations. In (c), we give our results under automatic settings and the results of competing methods.

Overview structure of our proposed FCNet: Two main components g and the colorization network f.

Dependencies and Installation

• Python >= 3.7 (Recommend to use Anaconda or Miniconda)
• PyTorch >= 1.7
• NVIDIA GPU + CUDA

1. Clone the repo

   git clone https://github.com/HyZhu39/FCNet.git

2. Install Dependencies

   cd FCNet
   pip install -r requirements.txt

Get Started

Test

We provide quick test code with the pretrained model.

1. Download this repo, as well as the pretrained models from Google Drive or Baidu Drive, and unzip.

2. Modify the paths to the test dataset and pretrained model in the following test scripts for configuration.

   ./scripts/demo_phase1.py
   ./scripts/demo_phase1_mul.py
   ./scripts/demo_phase2.py
   ./scripts/demo_phase3.py
   

3. Run test code for single reference image-guided colorization.

   python scripts/demo_phase1.py
   

4. Run test code for sample-guided colorization.

   python scripts/demo_phase2.py
   

5. Run test code for automatic colorization.

   python scripts/demo_phase3.py
   

6. Check out the results in ./results.

Train

1. Prepare datasets.

2. Modify the config file and train script.

   ./training/config.py
   ./scripts/train.py
   

3. Run training code.

Results

Result of *single- or multi-reference image-guided* colorization.

Result of single- or multi-reference sampling-guided colorization.

Result of automatic colorization.