Masked Regression (IJCV, 2020)

MR - Python Implementation

This repositery provides a python implementation of MR (Masked Regression). MR can efficiently synthesize facial expressions. The demo video for MR can be found here.

Masked Linear Regression for Learning Local Receptive Fields for Facial Expression Synthesis

Nazar Khan¹ · Arbish Akram¹ · Arif Mahmood² · Sania Ashraf¹· Kashif Murtaza¹

¹ Punjab University College of Information Technology (PUCIT), Lahore, Pakistan
² Department of Computer Science, Information Technology University (ITU), Lahore, Pakistan
International Journal of Computer Vision (IJCV), 2020

Usage

1. Download any Facial Expression Synthesis dataset

2. Create a folder structure as described here.

• Split images into training and test sets (e.g., 90%/10% for training and test, respectively).
• Crop all images to 128 x 128, where the faces are centered.

3. Training

To train MR:

$ python main.py --mode train --train_dataset_dir 'dataset/train/'  --image_size 128 --total_images 200 --input_ch 1 
                        --receptive_field 3 --lamda 0.4 

4. Test

To test MR:

$ python main.py --mode test --test_dataset_dir 'dataset/test/' --image_size 128 --total_images 20 --input_ch 1 
                        --receptive_field 3 

5. Test in the wild

To test MR:

$ python main.py --mode test_inthewild --test_dataset_dir 'dataset/inthewild/' --image_size 128 --total_images 20 --input_ch 1 
                        --receptive_field 3 

Results

Facial expression synthesis on sketches and animals

Facial expression synthesis on in the wild images

Citation

If this work is useful for your research, please cite our Paper:

@article{khan_mr_ijcv_2019,
author="Khan, Nazar and Akram, Arbish and Mahmood, Arif and Ashraf, Sania and Murtaza, Kashif", 
journal="International Journal of Computer Vision",
pages = "1433--1454",
title = "{Masked Linear Regression for Learning Local Receptive Fields for Facial Expression Synthesis}",
volume = "128",
year = "2020"
}