Deep Deinterlacing

This repository is an implementation of the paper Deep Deinterlacing by Disney research.

Results generated with the pretrained model which was trained for 10000 iterations. (MSE stands for mean squared error.)

Warning I am not one of the authors of the paper. The code in this repository was independently produced. It might contain errors and likely does not perfectly reflect what the authors did.

Warning In the original paper the interlacing is in the horizontal direction. This implementation works with interlacing in the vertical direction. This should not be too difficult to change however.

Setup

Installing requirements

Install the desired version of pytorch and torchvision as described on the pytorch website.

Run the following command to install all other dependencies.

pip install -r requirements.txt

Experiment tracking

To track experiments with Weights and Biases first install the wandb package.

pip install wandb

Then login with your Weigths and Biases authorization key using the following command

wandb login

To enable the experiment tracking set wandb to true in the model training config file.

Training a model

1. Make sure to download the dataset first and place it in the data folder.
2. Ensure all settings are as desired in the file training_config.yaml and disney_model.yaml
3. Run

python train_model.py

To train on the GPU add the argument --gpu 0 to train on GPU 0.

Further information

• Model definitions and training
• General utility functions

Model architecture