Open Access

Table 1

Reconstruction results: MSE, peak S/N, and SSIM metrics computed on preprocessed sky models for different diffusion settings.

Normalization Aggregation Metrics (± uncertainty)

MSE(×10−5) Peak S/N SSIM
γ = 1 single run 5.276 ± 0.138 38.031 ± 1.152 0.977 ± 0.002
mean 4.232 ± 0.333 42.243 ± 9.475 0.998 ± 0.001
median 3.941 ± 0.354 41.587 ± 9.156 0.998 ± 0.001
γ = 2 single run 1.492 ± 0.146 38.999 ± 0.097 0.830 ± 0.003
mean 1.189 ± 0.604 41.251 ± 0.206 0.853 ± 0.003
median 1.190 ± 0.605 42.148 ± 0.260 0.882 ± 0.003
γ = 10 single run 1.668 ± 0.171 36.784 ± 0.464 0.814 ± 0.004
mean 1.183 ± 0.592 38.008 ± 2.263 0.835 ± 0.005
median 1.185 ± 0.586 38.386 ± 2.550 0.862 ± 0.005
γ = 20 single run 2.264 ± 0.242 35.105 ± 0.793 0.892 ± 0.006
mean 1.431 ± 0.645 37.457 ± 4.474 0.929 ± 0.011
median 1.442 ± 0.643 37.651 ± 4.807 0.955 ± 0.008
γ = 30 single run 2.820 ± 0.313 32.944 ± 1.086 0.985 ± 0.002
mean 1.553 ± 0.750 34.844 ± 4.585 0.985 ± 0.008
median 1.569 ± 0.735 34.693 ± 4.906 0.988 ± 0.007

Notes. We varied the power root used in the normalization during training (Eq. (1)) and the aggregation method (mean, median over 20 runs, or single run). Single run signifies that the DDPM is run once for a given condition to get the predicted sky model. We evaluate performance on each run individually and then calculate the average score over all 20 runs. The error is calculated by taking the standard deviation over all outputs for the test set for aggregation methods, and over all runs to get a single run score. Italic font indicates the optimal metric value across all conducted experiments.

Current usage metrics show cumulative count of Article Views (full-text article views including HTML views, PDF and ePub downloads, according to the available data) and Abstracts Views on Vision4Press platform.

Data correspond to usage on the plateform after 2015. The current usage metrics is available 48-96 hours after online publication and is updated daily on week days.

Initial download of the metrics may take a while.