Unconstrained Open Vocabulary Image Classification: Zero-Shot Transfer from Text to Image via CLIP Inversion

Benchmark Model Rank Results
image-classification-on-imagenetDFN-5B H/14-378 + PrefixedIter Decoder#55Top 1 Accuracy: 88.21%
image-classification-on-imagenetSigLIP B/16 + PrefixedIter Decoder#412Top 1 Accuracy: 83.46%
open-vocabulary-image-classification-on-ovicSigLIP SO/14 + PrefixedIter Decoder (FT2)#1Prediction Score (mean of 3): 87.49
open-vocabulary-image-classification-on-ovicDFN-5B H/14-378 + PrefixedIter Decoder (FT2)#2Prediction Score (mean of 3): 87.08Overall Score: 87.13
open-vocabulary-image-classification-on-ovicDFN-5B H/14-378 + PrefixedIter Decoder (FT0)#3Prediction Score (mean of 3): 86.41Overall Score: 87.90
open-vocabulary-image-classification-on-ovic-1DFN-5B H/14-378 + PrefixedIter Decoder (FT0)#1Overall Score: 78.21Prediction Score: 79.18Top 1 Accuracy: 77.10
open-vocabulary-image-classification-on-ovic-1DFN-5B H/14-378 + PrefixedIter Decoder (FT2)#2Overall Score: 79.02Prediction Score: 80.13Top 1 Accuracy: 77.05
open-vocabulary-image-classification-on-ovic-1SigLIP B/16 + PrefixedIter Decoder (FT2)#3Prediction Score (mean of 3): 72.03