Fig. 4: Zero-shot remote sensing scene classification results.
From: Towards artificial general intelligence via a multimodal foundation model
a Zero-shot accuracies (%) on UCM with different unseen/seen class ratios. b Zero-shot accuracies (%) on AID with different unseen/seen class ratios. c Visualizations of “baseball field”. For (a) and (b), we report standard deviations in brackets over 25 random splits. Highest results in (a) and (b) are highlighted in bold.
