English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Plus annual

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Plus monthly

Global: Zendy Tools annual

Global: Zendy Tools monthly

Global: Zendy Free Plan

Synthetic data (SIM) drawn from simulators have emerged as a popularalternative for training models where acquiring annotated real-world images isdifficult. However, transferring models trained on synthetic images toreal-world applications can be challenging due to appearance disparities. Acommonly employed solution to counter this SIM2REAL gap is unsupervised domainadaptation, where models are trained using labeled SIM data and unlabeled REALdata. Mispredictions made by such SIM2REAL adapted models are often associatedwith miscalibration - stemming from overconfident predictions on real data. Inthis paper, we introduce AUGCAL, a simple training-time patch for unsupervisedadaptation that improves SIM2REAL adapted models by - (1) reducing overallmiscalibration, (2) reducing overconfidence in incorrect predictions and (3)improving confidence score reliability by better guiding misclassificationdetection - all while retaining or improving SIM2REAL performance. Given a baseSIM2REAL adaptation algorithm, at training time, AUGCAL involves replacingvanilla SIM images with strongly augmented views (AUG intervention) andadditionally optimizing for a training time calibration loss on augmented SIMpredictions (CAL intervention). We motivate AUGCAL using a brief analyticaljustification of how to reduce miscalibration on unlabeled REAL data. Throughour experiments, we empirically show the efficacy of AUGCAL across multipleadaptation methods, backbones, tasks and shifts.

AUGCAL: Improving Sim2Real Adaptation by Uncertainty Calibration on Augmented Synthetic Images