BayesCap: Bayesian Identity Cap for Calibrated Uncertainty in Frozen Neural Networks

Gradient-Based Uncertainty for Monocular Depth Estimation

“…We use the Eigen split [12] with the maximum depth set to 80 meters for evaluation. In [46], a small OOD evaluation is conducted with KITTI as in-distribution and Places365 [55] and India Driving [47] as OOD. In contrast to this evaluation, we suggest using two different in-distribution datasets and treating the OOD datasets as separate settings instead of combining them.…”

“…Another predictive method infers a prior distribution over the model output by log-likelihood maximization [39]. In [46], the prior output distribution is learned with a second model from a fixed and already trained image-to-image translation model. In contrast, we train a second decoder in a post hoc fashion to reconstruct the input image from the features of a depth encoder and thus learn its output distribution.…”

“…The main distinction is between data uncertainty (aleatoric), which results from noise in the data, and model uncertainty (epistemic), which is due to lack of knowledge and therefore addressable with more diverse training data [29]. Recent works [22,29,41,46] already target uncertainty estimation for depth estimation in the context of deep neural networks. Moreover, they evaluate whether the uncertainty corresponds to the actual error in the test data drawn from the same distribution as the training data.…”

Multi-Scale Aggregation Residual Channel Attention Fusion Network for Single Image Deraining

“…In the case of distribution P (J|I) in the above equation, recent studies have suggested that, in the presence of outliers and human factors, the residuals (the difference between predicted values and ground-truth values) of each pixel typically follow a heavy-tailed distribution [19], [33], [39]. Therefore, to make our model robust to various outliers, we model the heteroscedastic generalized Gaussian distribution to simulate the heavy-tailed distribution of the output.…”

“…The quality of results obtained from tests and measurements is crucial in practical vision-based measurement applications such as autonomous driving and medical imaging, and estimating uncertainty provides an effective means to assess outcome quality [16]- [18]. Well-calibrated uncertainty enables experts or devices to intervene in highly uncertain predictions, thereby enhancing decision-making reliability [19]- [21]. Therefore, researchers are encouraged to address the uncertainty in deep learning in order to enhance the reliability of measurement systems based on deep learning [16].…”