Multi Task Learning for simultaneous Semantic Segmentation, Instance Segmentation and Depth Estimation

Course Project for Deep Learning (Worcester Polytechnic Institute)

Work on progress!! Instance head development in "inseg" branch

This repository heavily depends on the work mentioned in the references section.

How to run:

1. The dataset should be downloaded from the following links and placed in a directory named cityscapes
   1. https://www.cityscapes-dataset.com/file-handling/?packageID=1
   2. https://www.cityscapes-dataset.com/file-handling/?packageID=3
   3. https://www.cityscapes-dataset.com/file-handling/?packageID=7
      
2. To run training script

Python train_cityscapes.py

3. To run inference script

Python inference_cityscapes.py

Methodology:

This code works with both NYUD and Cityscapes at the moment.

Current Results:

Cityscapes

NYUD

Team:

1. Bharath Kumar Ramesh Babu
2. Ghokulji Selvaraj
3. Durga Prakash Karuppannan
4. Krishna Sathwik Durgaraju

References

1. Real-Time Joint Semantic Segmentation and Depth Estimation Using Asymmetric Annotations
2. Semantic Instance Segmentation with a Discriminative Loss Function