Reproduce “Uncertainty Quantification in Detection Transformers: Object-Level Calibration and Image-Level Reliability” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Young-Jin Park, Carson Sobolewski, Navid Azizan
UQ-DETR evaluates predictions and filtering using Object-level Calibration Error (OCE), which penalizes both unreliable retained boxes and missed objects. Hungarian matching can leave unmatched predictions with low foreground confidence despite plausible boxes. Validation-set OCE selects predictions for ContrastiveConf, which contrasts retained and discarded confidence. On Cal-DETR/COCO, it correlates with per-image AP at 0.628, versus 0.547 for average retained confidence. This is correlation, not improved accuracy.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.