Spatial-TemporalSuper-ResolutionofSatellite ImageryviaConditionalPixelSynthesis

We provide two versions of the image feature mapper. In version "EAD", we use two3 3 convolutional layers with stride= 2 in FE, a single self-attention module inFA and two3 3 transposed convolutional layers withstride=2inFD. We then extract 4 images for each house (NAIP in 2016, NAIP in 2018, Sentinel-2 in 2016, and Sentinel-2 in2018). We reserve 237 locations as the testing set and the remaining 1515 locations as the training set. Vigilance Tests Each HIT contains a vigilance image for which object counts are sufficiently unambiguous.