Adithya Gaurav Singh’s Post

A lot of top Autonomous Driving companies have secretly started shifting to Multi-Task Learning Neural Networks to streamline the data processing from multiple cameras installed on a Vehicle. Tesla has named this methodology as "Hydranet" What if you could execute Autonomous Driving tasks like Segmentation, Object Detection, Depth Estimation etc, using a single Neural Network in a single forward pass instead of having a separate model for each task? This is the essence of Multi-Task Learning. I finished my Fall semester as Robotics Masters student at University of Maryland, College Park with this project to implement a Multi-Task neural network. It processes an input image and predicts a Segmentation Mask and Depth Map, in a single model in a single forward pass. These outputs when fused together we can obtain exciting results like 3D Segmentation at a very low cost of resources. I thank Jérémy Cohen and Adrian Rosebrock for their guidance for this project through their course content at Think Autonomous (https://lnkd.in/gGUEjRhJ) and PyImageSearch University ( https://lnkd.in/g_C5WBST). Project : https://lnkd.in/ggv76yWc #MultiTaskLearning #Hydranet #autonomousdriving #computervision #deeplearning