https://github.com/nikhilroxtomar/multiclass-image-segmentation-using-unetr-in-tensorflow

Multiclass Image Segmentation on Human Face Segmentation in TensorFlow
https://github.com/nikhilroxtomar/multiclass-image-segmentation-using-unetr-in-tensorflow

multiclass-segmentation tensorflow unet unet-transformer unetr unetr-segmentation

Last synced: 5 months ago
JSON representation

Multiclass Image Segmentation on Human Face Segmentation in TensorFlow

• Host: GitHub
• URL: https://github.com/nikhilroxtomar/multiclass-image-segmentation-using-unetr-in-tensorflow
• Owner: nikhilroxtomar
• Created: 2024-02-02T12:37:40.000Z (almost 2 years ago)
• Default Branch: main
• Last Pushed: 2024-02-02T12:49:07.000Z (almost 2 years ago)
• Last Synced: 2024-11-16T06:28:02.746Z (about 1 year ago)
• Topics: multiclass-segmentation, tensorflow, unet, unet-transformer, unetr, unetr-segmentation
• Language: Python
• Homepage:
• Size: 2.5 MB
• Stars: 4
• Watchers: 2
• Forks: 2
• Open Issues: 0
• Metadata Files:
  • Readme: README.md

Awesome Lists containing this project

README

          
# Multiclass-Image-Segmentation-using-UNETR-in-TensorFlow

This GitHub repository demonstrates the utilization of UNETR for multiclass image segmentation on the Landmark Guided Face Parsing dataset (LaPa) dataset.

.

## Architecture

| ![The block diagram of the Original UNETR model.](figures/unetr_architecture.png) |

| :--: |

| *The block diagram of the Original UNETR model.* |

## Dataset

LaPa stands for Landmark guided face Parsing dataset (LaPa). It is a large-scale dataset for human face parsing. It consists of more than 22,000 facial images with abundant variations in expression, pose and occlusion, and each image of LaPa is provided with a 11-category pixel-level label map and 106-point landmarks.


 


Download the dataset: [Landmark guided face Parsing dataset (LaPa)](https://drive.google.com/file/d/1XOBoRGSraP50_pS1YPB8_i8Wmw_5L-NG/view?usp=sharing)


 


For more: [LaPa-Dataset for face parsing](https://github.com/jd-opensource/lapa-dataset)

## Results

The sequence in the images below is `Input Image`, `Ground Truth` and `Prediction`. 
 


| ![](results/115769026_1.png) |

| :--: |

| ![](results/101972691_1.png) |

| ![](results/141022532_0.png) |

| ![](results/218281108_2.png) |

| ![](results/1048174932_2.png) |

## How to improve

- Train on more epochs.

- Increase the input image resolution.

- Apply data augmentation.

## Contact

For more follow me on:

-  YouTube 

-  Facebook 

-  Twitter 

-  Instagram 

-  Telegram