Pritom Gogoi

Renningen, DE

Proposed a Low-Rank and Diagonal (LRD) covariance model for probabilistic online map generation in autonomous driving. Integrated uncertainty into downstream trajectory prediction via novel encoding schemes and confidence score-based feature modulation, leading to SOTA results on the nuScenes dataset.

Stuttgart, DE

Worked under Fabian Kögel (supervised by Dr. Andreas Bulling) on extending MTEX, an explainability framework, by implementing input-attribution methods for multimodal transformer models like ViLT. Integrated AttnLRP, integrated gradients and input × gradients into the framework and ran perturbation experiments.

Darmstadt, DE

Extended an in-house human pose estimation framework to predict relative depth values for pose joints. Developed a pipeline to enrich human pose datasets with depth annotations by leveraging DepthAnythingV2, a monocular depth estimation model.

München, DE

Developed predictive ML models to forecast client contact frequency and timing using a rich dataset with 50+ attributes. Emphasized explainable predictions to reveal key features influencing outcomes, aiming for impactful business insights.

Stuttgart, DE

• Bayesflow: Worked under Marvin Schmitt to support the development of the open-source package Bayesflow, which enables efficient Bayesian inference with deep learning.
• ARTSM: Worked under Christian Pfaendner on an open-source Python package for backmapping of coarse-grained molecular structures into atomistic resolution using machine learning.

Stuttgart, DE

Worked under Dr. Mihai Bâce on a deep learning pipeline for generating human eye images using image-to-image translation. Used models like GroundingDINO for object detection and Segment Anything (SAM) for semantic segmentation.

Guwahati, IN

Built open-source implementations of hand pose estimation models, including the Attention-based 2D Hand Pose Estimation model by Santavas et al. and the Convolutional Pose Machines paper by Wei et al. Both used the TensorFlow framework.

Stuttgart, DE · Grade: 1.5

• Master’s Thesis: Uncertainty-Aware Map Generation for Motion Prediction in Autonomous Driving.
• Research Project — Diffusion-based Gaze Estimation: A gaze-estimation-from-a-distance model using conditional score-based diffusion to model the temporal coordination between gaze, head and body orientations.
• Relevant Coursework: Deep Learning, Machine Learning, Matrix Computations in Signal Processing, Computer Architecture & Organization, Operating Systems, GPU Programming, Signals and Systems.

Guwahati, IN · Percentage: 84.13

• Bachelor’s Thesis — Image Processing Based Object Tracking Turret: A servo turret prototype controlled by a Raspberry Pi 4 that detects and tracks COCO objects using a MobileNet model, with motion controlled by a PID controller.
• Relevant Coursework: Multivariate Calculus, Linear Algebra, Probability & Statistics, Digital System Design, Data Structures & Algorithms, Embedded Systems Development, Digital Image Processing.

Uncertainty Matters: Structured Probabilistic Online Mapping for Motion Prediction in Autonomous Driving

P. Gogoi, F. Janjoš, B. Yang and A. Look

arXiv:2603.20076 · planned for ICRA 2027 — view on arXiv

An UAV-Assisted Multi-sensor Based Smart Parking System

P. Gogoi, J. Dutta, R. Matam, M. Mukherjee

IEEE Intl. Conference on Computer Communications (INFOCOM) · Ottawa, Canada, Jul 2020 — view paper

ART-SM: Boosting Fragment-Based Backmapping by Machine Learning

C. Pfaendner, V. Korn, P. Gogoi, B. Unger and K. Pluhackova

Journal of Chemical Theory and Computation (2025)

A GPU-accelerated JPEG encoder in C++ that improves image encoding/decoding throughput by leveraging NVIDIA CUDA for high-performance parallel processing.

Used the Gazepoint GP3 eyetracker for precise gaze-pattern analysis to study the cognitive mechanisms underlying real-time language comprehension.

Designed a 16-bit CPU with a custom instruction set based on the self-taught nand2tetris curriculum, and wrote a Python assembler for its instruction set.

Implementations of Computer Vision and Deep Learning papers — CGAN, Neural Style Transfer, Convolutional Pose Machines and more.