HydraNets-for-Multitask-Learning

This repository contains the Jupyter Notebooks for HydraNets for multi-task learning for machine learning and and computer vision tasks. These tasks are :

Machine Learning

• Classification
• Regression

Computer Vision

• Image Segmentation
• Depth Map

The machine learning tasks are performed on the UTKFace dataset while the computer vision tasks are performed on the KITTI and NYUD datasets used in the research paper Real-Time Joint Semantic Segmentation and Depth Estimation Using Asymmetric Annotations

Note : The TensorFlow version of the computer vision tasks (KITTI and NYUD folders) are W.I.P. This is because the PyTorch versions of these notebooks uses the pretrained model checkpoints and weights provided by the authors of the paper which cannot be used for TensorFlow. The PyTorch trained model can be converted to ONNX format to be loaded using TensorFlow for making inference but for the KITTI dataset this method causes shapes mismatch and hasn't yet been implemented and explored for NYDU dataset.