Computer Science Seminar
Date: Friday, 24 May 2024
Time: 12:00 PM
Venue: Lecture Hall 5
Safe and Efficient Semantic Segmentation for Road Scenes
Girish Varma
IIIT Hyderabad.
24-05-24

---

Abstract

Semantic Segmentation is a fundamental primitive for several tasks in robotics and autonomous navigation. Real world deployment of such systems requires models that are highly efficient as well as the predictions should have a very high degree of robustness to unstructured traffic, weather conditions, and should prevent unsafe mispredictions. We develop datasets tailored to unstructured driving conditions prevailing in India, that allows development of robust models [WACV'19,'24]. We build universal models that can predict well in different domains using the same trained weights [ICCV'19]. We also use methods from graph theory to come up with efficient CNN layers that result in fast models [ECCV'18, CVPRW'19].

Speaker Bio: Girish Varma is a faculty at the Center for Security, Theory & Algorithms Research (CSTAR) and the Machine Learning Lab at IIIT Hyderabad, broadly interested in theoretical and applied computer science problems. In the Transportation domain, he has worked on Indian Drive Scenes Segmentation Datasets, Routing Problems etc.