Home

Awesome

#ace-Matting-using-Unet Text-Detection-Model Text Detection Model trained with Precision Score of 98% on Test Set of 200 images.

Description

Potrait Matting Model with 90% accuracy build on fine tuning vgg_unet using https://github.com/switchablenorms/CelebAMask-HQ Dataset. This model is made using amazing package made by Divam Gupta : https://divamgupta.com. Link to his repo https://github.com/divamgupta/image-segmentation-keras

Setup

  1. create a viruatual environment and activate it
  2. install required libs
   apt-get install -y libsm6 libxext6 libxrender-dev
   pip install opencv-python
  1. install keras-segmenatation pacakge using
pip install keras-segmentation

  1. Downlaod Celeb Dataset

  2. Edit the path to the dataset in preprocessing.py and run it.

python preprocessing.py
  1. Run the traning.py file using
python traning.py
  1. Test the image using inference-image.py

Samples

3 images from the datset - 6 images from the internet

Disclaimer - I do not own/clicked any of the images below.

<br /> <p><img align="left" width="425" height="425" src="https://github.com/Divyam10/Face-Matting-using-Unet/blob/master/results/0.jpg"> <img align="right" width="425" height="425" src="https://github.com/Divyam10/Face-Matting-using-Unet/blob/master/results/0.png"></p> <br /> <p><img align="left" width="425" height="425" src="https://github.com/Divyam10/Face-Matting-using-Unet/blob/master/results/1.jpg"> <img align="right" width="425" height="425" src="https://github.com/Divyam10/Face-Matting-using-Unet/blob/master/results/1.png"></p> <br /> <p><img align="left" width="425" height="425" src="https://github.com/Divyam10/Face-Matting-using-Unet/blob/master/results/2.jpg"> <img align="right" width="425" height="425" src="https://github.com/Divyam10/Face-Matting-using-Unet/blob/master/results/2.png"></p> <br /> <p><img align="left" width="425" height="425" src="https://github.com/Divyam10/Face-Matting-using-Unet/blob/master/results/4.jpg"> <img align="right" width="425" height="425" src="https://github.com/Divyam10/Face-Matting-using-Unet/blob/master/results/4.png"></p> <br /> <p><img align="left" width="425" height="425" src="https://github.com/Divyam10/Face-Matting-using-Unet/blob/master/results/5.jpg"> <img align="right" width="425" height="425" src="https://github.com/Divyam10/Face-Matting-using-Unet/blob/master/results/5.png"></p> <br /> <p><img align="left" width="425" height="425" src="https://github.com/Divyam10/Face-Matting-using-Unet/blob/master/results/6.jpg"> <img align="right" width="425" height="425" src="https://github.com/Divyam10/Face-Matting-using-Unet/blob/master/results/6.png"></p> <br /> <p><img align="left" width="425" height="425" src="https://github.com/Divyam10/Face-Matting-using-Unet/blob/master/results/7.jpg"> <img align="right" width="425" height="425" src="https://github.com/Divyam10/Face-Matting-using-Unet/blob/master/results/7.png"></p> <br /> <p><img align="left" width="425" height="425" src="https://github.com/Divyam10/Face-Matting-using-Unet/blob/master/results/8.jpg"> <img align="right" width="425" height="425" src="https://github.com/Divyam10/Face-Matting-using-Unet/blob/master/results/8.png"></p> <br />

Extentions to the project

  1. Face attritube changing like color of the hair,skin
  2. potrait regeneration using Gans, regenerating new attributes like long hair, beard, etc
  3. Traninig on dataset with lesser quality to apply matting on non-hd images