Papers
5
Total Citations
56
H-Index
4
About
Dimity Miller is a computer vision and robotics researcher whose work centers on uncertainty quantification, probabilistic detection, and the robustness of deep learning systems in real-world deployment. Her research addresses a critical challenge in modern AI: ensuring that object detectors not only perform well under controlled conditions but also behave reliably and transparently when encountering novel or ambiguous scenarios. Miller's most influential contribution, "What's in the Black Box?" (2022, 22 citations), provides a rare mechanistic breakdown of why object detectors fail, identifying five distinct internal failure modes — offering practitioners actionable diagnostic insight rather than surface-level performance metrics. Her foundational work on Probabilistic Object Detection (2020, 16 citations) introduced a rigorous framework and evaluation metric for quantifying spatial and semantic uncertainty in detections, helping establish a new subfield. She has also pioneered the application of Dropout Sampling to object detection (2018, 9 citations) and developed GMM-Det, a real-time method for identifying open-set errors using epistemic uncertainty. More recently, her work extends uncertainty awareness to lidar-based place recognition, broadening her impact into autonomous navigation. Across her portfolio, Miller consistently bridges theoretical rigor with practical deployability, making her research essential reading for those working on trustworthy autonomous systems.
Research Focus
Key Achievements
Top Papers
- 1
- 2Probabilistic Object Detection: Definition and Evaluation16 citations · 2020
- 3Dropout Sampling for Robust Object Detection in Open-Set Conditions9 citations · 2018
- 4Uncertainty-Aware Lidar Place Recognition in Novel Environments5 citations · 2023
- 5Uncertainty for Identifying Open-Set Errors in Visual Object Detection4 citations · 2021