flip is one of our standard CV augments, but as was noted in the iNat forums, some molluscs are not horizontally symmetrical. we get a few percent accuracy boost for using flip, but can we quantify if we're actually underperforming on pairs of mollusk species that are primarily distinguishable via spiral direction?
flip is one of our standard CV augments, but as was noted in the iNat forums, some molluscs are not horizontally symmetrical. we get a few percent accuracy boost for using flip, but can we quantify if we're actually underperforming on pairs of mollusk species that are primarily distinguishable via spiral direction?
see https://forum.inaturalist.org/t/how-are-photos-selected-for-cv-training/42403/20 for a discussion