Publications · Manpa Barman

Figure from Rare Object Retrieval for Autonomous Driving

Master's ThesisMercedes-Benz R&DUniversität Stuttgart

Under review

Rare Object Retrieval for Autonomous Driving

F. Embacher*, M. Barman*, J. Uhrig, M. Enzweiler

Submitted to WACV, 2026

*Equal contribution

object retrievalvision-language modelsobject detectionDETRDINO

Universität StuttgartComputational Cognitive Science Group

Automating Data Integration and Publishing for Neuroimaging via LSLAutoBIDS

M. Barman, J. Range, B. Ehinger

Aperture Neuro, Vol. 6, Issue SI 1, 2026

Presented at deRSE 2026 (presenting author)

DOI Code

EEGeye trackingBIDSopen scienceneuroimaging

Project details

Open-source Python package built with the Computational Cognitive Science group (Prof. Benedikt Ehinger, Universität Stuttgart) to standardize, version and archive the EEG and eye-tracking data produced by neuroscience experiments — a large, unwieldy corpus that’s otherwise tedious to manage.

Data standardization: converts raw recordings into BIDS-compliant datasets
Data versioning: uses Datalad for version control
Data archiving: uses Dataverse for long-term archiving

Figure from From Monitoring to Forecasting: An Integrated Visitor Insights Tool for Protected Area Management

DSSGx MunichLMU München

Under review

From Monitoring to Forecasting: An Integrated Visitor Insights Tool for Protected Area Management

M. Mangold, M. Barman, A. Garove, P. Ferreira, A. Younas, J. Ostheimer, J. von Bartenwerffer, M. Heurich

Journal of Outdoor Recreation and Tourism

Code

time-series forecastingdata pipelinesvisitor monitoringdashboarding

Project details

Built as part of Data Science for Social Good Munich (DSSGx) 2024, with an interdisciplinary team of economists, social scientists and computer scientists, to harmonize tourism and nature protection in the Bavarian Forest National Park.

Data pipeline harmonizing sensor and survey data from multiple sources
Predictive model forecasting weekly visitor traffic
Dashboard visualizing park insights and forecasted traffic for park management
Technical documentation of the pipeline, model and suggested future work

Universität StuttgartCollaborative Artificial Intelligence Group

CryptoQA: A Large-scale Question-answering Dataset for AI-assisted Cryptography

M. Elfares, P. Reisert, T. Dietz, M. Barman, A. Zaki, R. Küsters, A. Bulling

arXiv preprint arXiv:2512.02625, 2025

arXiv Code

LLM fine-tuningcryptographyquestion answeringdataset curation

Project details

Built an LLM-based assistant for cryptography problems, which need a blend of natural-language understanding and multistep mathematical reasoning that general-purpose LLMs handle poorly. Curated CryptoQA, a dataset drawn from academic textbooks and resources, and fine-tuned four CryptoLLM variants — starting from both a math-tuned and a general-purpose base model, with single-stage and two-stage (math dataset, then CryptoQA) fine-tuning — then ran a qualitative behavioural analysis of base vs. fine-tuned models.

CryptoQA: a public dataset of cryptography questions
Four fine-tuned CryptoLLM model variants
Two-stage fine-tuning recipe (general math → cryptography)
Qualitative analysis of base vs. fine-tuned reasoning behaviour

Bachelor ThesisAssam Engineering College

Object Detection and Tracking Turret based on Cascade Classifiers and Single Shot Detectors

P. Gogoi, M. Barman, M. Deka, U. Rajkonwar, R. Moudgollya

International Conference on Computational Performance Evaluation (ComPE), Shillong, India, 2020

object detectionobject trackingembedded systemsMobileNet

Deep Learning Based Flood Mapping: A Case Study from a Flood-vulnerable State in Northeastern India

M. Barman, P. Gogoi, S. Goswami, H. Nagendra

Conference abstract, American Geophysical Union (AGU) Fall Meeting, Chicago, USA, 2022

flood mappingremote sensingdeep learning