andy580/imageColorization

Prerequisites

1. Install conda.
2. Create an environment

conda create --name virtual python=3.7
conda activate virtual

3. Install dependencies

• A NVIDIA® GPU card is needed check CUDA® Enabled GPUs
• For RTX 30 series cards, check Build tensorflow from source to install tensorflow for CUDA® 11.1

pip install numpy pandas matplotlib scikit-image
pip install opencv-python==3.4.11.41 opencv-contrib-python==3.4.11.41
conda install -c anaconda cudatoolkit==10.1 cudnn
pip install tensorflow-gpu scikit-learn

Getting Started

1. Clone this repo:

git clone https://github.com/r444li/SYDE671-final_project-image_colorization.git

2. Download dataset (~18GB).
3. Download files of bounding boxes.
4. Create 4 directories for training images and annotations, validating images and annotations
5. Create 1 directory for checkpoints
6. Update paths in file_paths.py
7. Put all the images downloaded in the directory of training images
8. Put all the files of bounding boxes downloaded in the direcotry of train annotations
9. Run clean_images.py

Training Full Model

1. Set batch_size, epochs, img_size in Train_Full_ImageColorizationNetwork_reduced.py(Line 17-19)
2. Run Train_Full_ImageColorizationNetwork_reduced.py

Training Instance Model

1. Set batch_size, epochs, img_size in Train_Instance_ImageColorizationNetwork_reduced.py(Line 17-19)
2. Run Train_Instance_ImageColorizationNetwork_reduced.py

Training Fusion Model

• Before train fusion model, there must be at least one file of weights for each of full and instance models in the directory of checkpoints

1. Choose one weights file for each of full and instance models in the directory of checkpoints.
2. Update full_imagecolorization_weights and instance_imagecolorization_weights in file_paths.py
3. Set epochs, img_size in fusion_train_reduced.py(Line 22-23)
4. Run fusion_train_reduced.py

View the results of models

1. Choose one weights file in the directory of checkpoints.
2. Create 1 directory for outputs
3. Update recent_weights and output_path in file_paths.py to the chosed file.
4. Update train_generator and valid_generator in test.py(start at Line 30) to the definition in the training program of chosed model.
5. Run test.py