Follow the steps below to set up your environment, prepare your datasets, and run the training and testing scripts.

Ensure your movie files are placed in a directory named movies. The movie_frame_parser.py script will extract frames from these movies.

To extract frames from the movies:

1. Ensure the movies directory is in the same directory as movie_frame_parser.py.

2. Run the script to extract frames:

   python movie_frame_parser.py

   This script will read each movie file from the movies directory, extract every 72nd frame, and save them in a directory structure like frames/frames_<movie_filename>.

Prepare your dataset directories and ensure they are structured correctly. Here are the required datasets:

• Cartoon Dataset: Extracted frames from movies like "Your Name" and "Weathering with You".
• Landscape Dataset: High-quality landscape images (e.g., LHQ dataset at 256 resolution).

The datasets should be placed in the data directory with the following structure:

data/
  (input_images)/
  (name_of_movie_frames)/
  test-v2/

Download the vgg19.npy file from the following link: VGG19 Weights

Place the downloaded vgg19.npy file in the same directory as vgg19.py.

To start the training process with default parameters, use the following command:

python main.py --mode train

To customize the training parameters, you can specify additional arguments:

python main.py --mode train --image_size 256 --batch_size 16 --pre_train_iter 2000 --iter 20000 --learning_rate 1e-4 --save_dir 'saved_models' --train_out_dir 'train_output'

• image_size: Size of the input images (default is 256).
• batch_size: Number of images per batch (default is 16).
• pre_train_iter: Number of pre-training iterations (default is 5000).
• iter: Number of training iterations (default is 50000).
• learning_rate: Learning rate for the optimizer (default is 1e-4).
• save_dir: Directory to save the trained models.
• train_out_dir: Directory to save training output images.

To test the trained model, use the following command:

python main.py --mode test

Ensure that the test-v2 dataset is prepared in the data directory and contains the test images.

This script extracts frames from movie files:

• File and Directory Setup: Identifies the root directory and movie file names.
• Video Capture: Reads the movie files using cv2.VideoCapture.
• Frame Extraction: Saves every 72nd frame as an image.
• Main Function: Iterates through all movie files and extracts frames.

This script handles various preprocessing tasks:

• Removing Small Pictures: Removes images smaller than 92x92 pixels.
• Increasing Resolution: Resizes images to 256x256 pixels.
• Arranging Data: Crops and resizes images to 256x256 pixels, maintaining aspect ratios.

This script sets up the VGG19 model:

• Loading Pre-Trained Weights: Loads weights from vgg19.npy.
• Building the Model: Constructs the VGG19 layers and sets up feature extraction.

This script contains the main training and testing procedures:

• Argument Parsing: Sets up command-line arguments for configuration.
• CartoonGAN Class: Initializes parameters, sets up data, builds the model, and defines training/testing procedures.
• Pre-Training: Trains the generator using VGG19 perceptual loss before adversarial training.
• Training: Alternates between updating the generator and discriminator using adversarial loss.
• Testing: Loads the trained model and generates images for the test dataset.