01
Geospatial AI
Foundation models, multimodal benchmarks, and agentic models for satellite imagery, weather, and air-quality forecasting.
Remote sensing · Climate AI · Foundation models
Research Scientist · Computer Vision
Muhammad Akhtar Munir
PhD Computer Science
I am a Research Scientist in Computer Vision at MBZUAI, developing reliable multimodal AI systems with a focus on vision-language models, foundation models, model calibration, and tool-augmented agents.
My work covers model evaluation and adaptation, robust prediction under domain shift, and agentic visual reasoning, with applications in object detection, geospatial intelligence, and environmental modeling. My research has appeared at CVPR, ICCV, NeurIPS, ICLR, ECCV, and TPAMI, including Highlight papers and Best Paper awards.
Focus
Reliable AI · Real-world visual intelligence · Geospatial AI
Research
AI agents · Vision-language models · Foundation models
Published at
CVPR · ICCV · NeurIPS · ICLR · ECCV · TPAMI
Research
Research interests
My work is organized around three connected areas: building capable models, evaluating them rigorously, and making their confidence useful in practice.
02
Reliable multimodal learning
Evaluation and adaptation methods for vision-language models, with emphasis on calibration, uncertainty, and behavior under distribution shift.
Model calibration · Domain shift · VLM adaptation
03
Agentic models
Tool-augmented models that reason over visual evidence and can be evaluated on realistic, multi-step tasks.
Tool use · Visual reasoning · Benchmarking
Selected work
Representative projects
A selection illustrating the progression from calibrated perception to multimodal and agentic Earth-intelligence models.
ThinkGeo
A benchmark for evaluating whether tool-augmented agents can solve complex, multi-step remote-sensing tasks using visual evidence and external tools.
Publications
Selected publications
Full author lists and links are retained below. For citation counts and the complete record, see Google Scholar ↗.
16 publications
2026
CVPR 2026
Towards Calibrating Prompt Tuning of Vision-Language Models
2026
MONTI-CVPR · Best Paper
ThinkGeo: Evaluating Tool-Augmented Agents for Remote Sensing Tasks
2026
ECCV 2026
OpenEarthAgent: A Unified Framework for Tool-Augmented Geospatial Agents
2026
ICLR 2026
TerraFM: A Scalable Foundation Model for Unified Multisensor Earth Observation
2026
npj CleanAir
Synergistic Neural Forecasting of Air Pollution with Stochastic Sampling
2025
ICCV 2025 · Highlight
GEOBench-VLM: Benchmarking Vision-Language Models for Geospatial Tasks
2025
TerraBytes-ICML · Best Paper
AirCast: Improving Air Pollution Forecasting Through Multi-Variable Data Alignment
2025
CVPR 2025 · Highlight
O-TPT: Orthogonality Constraints for Calibrating Test-time Prompt Tuning in Vision-Language Models
2025
CVPR 2025
EarthDial: Turning Multi-sensory Earth Observations to Interactive Dialogues
2024
CCAI-NeurIPS 2024
Efficient Localized Adaptation of Neural Weather Forecasting: A Case Study in the MENA Region
2024
CVPR 2024
Improving Single Domain-Generalized Object Detection: A Focus on Diversification and Alignment
2023
NeurIPS 2023
Cal-DETR: Calibrated Detection Transformer
2023
IEEE TPAMI 2023
Domain Adaptive Object Detection via Balancing between Self-Training and Adversarial Learning
2023
CVPR 2023
Bridging Precision and Confidence: A Train-Time Loss for Calibrating Object Detection
2022
NeurIPS 2022
Towards Improving Calibration in Object Detection under Domain Shift
2021
NeurIPS 2021
SSAL: Synergizing between Self-Training and Adversarial Learning for Domain Adaptive Object Detection
People
Collaborators and mentees
I have been fortunate to work with students and early-career researchers across remote sensing, multimodal learning, calibration, and climate AI.
Advisors and senior collaborators: Salman Khan, Muhammad Haris Khan, Saquib Sarfraz, Fahad Shahbaz Khan, and Mohsen Ali.
Muhammad Sohail Danish
Remote-sensing benchmarks and foundation models
Akashah Shabbir
Agentic pipelines for remote sensing
Vishal Nedungadi
Air-pollution forecasting and climate AI
Ashshak Sharifdeen
Calibration for vision-language models
Yohan Abeysinghe
Generative methods for climate modeling
Syed Roshaan Ali Shah
Remote-sensing benchmarks and foundation models
Muhammad Umer Sheikh
Tool-augmented geospatial agents
Service & honors
Research community
Selected recognition, reviewing, teaching, and conference activity.
Recognition
- Best Paper — ICML TerraBytes Workshop, 2025
- Best Paper — CVPR MONTI Workshop, 2026
- Highlight Papers — CVPR and ICCV, 2025
- Outstanding Reviewer — CVPR, 2025
- NeurIPS Scholar Award — 2022 and 2023
- CVPR DEI Award — 2023
- UAE AI Award finalist — EarthDial, Scientific Research category, 2025
Professional service
- Reviewer: CVPR, ICCV, NeurIPS, ICLR, ECCV, EMNLP, TPAMI, and IJCV
- Guest lecturer: CV805 — Life-long Learning Agents for Vision, MBZUAI
- Invited speaker: University of Central Florida
- Conference presentations: CVPR, ICCV, and NeurIPS
- Research fellowship: Information Technology University, Pakistan
Contact
Research conversations and collaboration
I welcome conversations about reliable multimodal AI, geospatial intelligence, vision-language models, and related research collaborations.