Algorithm 1 SSAL: Training procedure with Uncertainty Guided Pseudo Labels(UGPL) and UncertaintyGuidedTiles(UGT) Input: Set of labeled data, Ds, and unlabeled data Dt, uncertainty and detection consistency thresholdsκ1 =0.5&κ2 =N/2Output:DomainadaptedtrainedmodelG 1: TrainthemodelGs,usinglabeleddata,Ds.Eq. Although we set both thresholds at 0.5, we find that our method is relatively robust to these hyperparameters. For instance, upon varying theκ1 by 0.1 unit in both directions, the maximum drop in mAP score is 0.6% (Tab. Calibration is measured using ECE score. Figure 1 shows more qualitative results for source-only, EPM [3], and our method.