This repository contains the trained neural network (TensorFlow) models, associated with the paper "MTJND: Multi-Task Deep Learning Framework for Improved JND Prediction" Published in IEEE ICIP 2023.

• Tensorflow
• FFmpeg

Our evaluation is conducted on VideoSet and MCL-JCI datasets.

Our pretrained models are capable of predicting JND values, and they can also be employed for training on a custom dataset.

- rootdir/
     - train/
         - img#1
         - img#2
         - ...
         - JND-Levels.txt (a file containing the 3 JND levels per image: first column for the first JND, second column for the second JND, and third column for the third JND level)
     - valid/
         - img#1
         - img#2
         - ...
         - JND-Levels.txt (a file containing the 3 JND levels per image: first column for the first JND, second column for the second JND, and third column for the third JND level)
     - test/
         - img#1
         - img#2
         - ...

For prediction with MT_3LJND or MT_3LJND_VA, the following commands can be used.

python3 MT_3LJND.py test --data_dir "Path-to-the-folder-containing-train,valid,and-test-subfolders/" --model_weights_path "Path-to-the-pretrained-model/model-name.h5" --result_path "Path-to-save-test-results/result.csv"

For prediction with MT_1LJND_VA, the following commands can be used.

python3 MT_1LJND_VA.py test --data_dir "Path-to-the-folder-containing-train,valid,and-test-subfolders/" --model_weights_path "Path-to-the-pretrained-model" --jnd_column int --result_path "Path-to-save-test-results/result.csv"

For more information on the implemented code and training process, refer to the paper's GitHub page.