Ekagra Gupta

Currently working on benchmarking a defect-classification task using DeiT vision transformer for high-quality industrial battery images. Streamlined dataset preparation by auto-labeling defects in high-resolution manufacturing component images using YOLOv5, SAM, and SAM2 for bounding box and mask generation, ensuring high-quality annotations and improved workflow efficiency.

During my internship, I applied data corruptions and instance manipulations to the nuScenes dataset to improve autonomous driving models. I evaluated these modifications using advanced object detection models like StreamPETR, FAR3D, and FIERY. The goal was to analyze model behavior under various scene- and instance-specific conditions, identify critical corner cases, and enhance the robustness of autonomous driving perception systems.

As a working student in the (e)Powertrain department at DaimlerTrucks AG, Stuttgart, I improved MoLaSim, an in-house engine mount analysis tool, by enhancing its capabilities and UI through VBA scripting. I also optimized data and model exports in Simpack, a multi-body simulation tool, by developing custom Python scripts, boosting efficiency and data accuracy. These contributions streamlined workflows and enhanced the team's analytical capabilities.

I collaborated on the implementation of a Python-based dashboard for transport route monitoring using GoogleMaps API, expanding its functionalities. Additionally, I independently developed an innovative approach for route planning, contributing to improved logistics operations.

As a part of collegiate team, my responsibilities included estimating vehicle poses using perception data, enhancing the already developed vehicle's localization system with C++ and ROS2, and mapping through the landmark based pose-graph method. This collaborative effort was instrumental in advancing our team's capabilities in autonomous driving technology.

Specialization: Smart Systems