. Author manuscript; available in PMC: 2024 Aug 14.

Published in final edited form as: Proc Conf Empir Methods Nat Lang Process. 2022 Dec;2022:3876–3887. doi: 10.18653/v1/2022.emnlp-main.256

Table 1:

Results of zero-shot image classification tasks on four datasets. We take an additional prompt ensemble version of each method (with subscript _ENS). We take the mean and standard deviation (STD) of accuracy (ACC) in five runs considering the randomness of prompt generation process. Best scores across a dataset are in bold.

ACC(STD)	CheXpert-5x200	MIMIC-5x200	COVID	RSNA
CLIP	0.2016(0.01)	0.1918(0.01)	0.5069(0.03)	0.4989(0.01)
CLIP_ENS	0.2036(0.01)	0.2254(0.01)	0.5090(<0.01)	0.5055(0.01)
ConVIRT	0.4188(0.01)	0.4018(0.01)	0.5184(0.01)	0.4731(0.05)
ConVIRT_ENS	0.4224(0.02)	0.4010(0.02)	0.6647(0.05)	0.4647(0.08)
GLoRIA	0.4328(0.01)	0.3306(0.01)	0.7090(0.04)	0.5808(0.08)
GLoRIA_ENS	0.4210(0.03)	0.3382(0.01)	0.5702(0.06)	0.4752(0.06)
`MedCLIP`-ResNet	0.5476(0.01)	0.5022(0.02)	0.8472(<0.01)	0.7418(<0.01)
`MedCLIP`-ResNet_ENS	0.5712(<0.01)	0.5430(<0.01)	0.8369(<0.01)	0.7584(<0.01)
`MedCLIP`-ViT	0.5942(<0.01)	0.5006(<0.01)	0.8013(<0.01)	0.7447(0.01)
`MedCLIP`-ViT_ENS	0.5942(<0.01)	0.5024(<0.01)	0.7943(<0.01)	0.7682(<0.01)