AttriPrompt: Dynamic Prompt Composition Learning for CLIP

Benchmark Model Rank Results
prompt-engineering-on-caltech-101AttriPrompt–Harmonic mean: 96.99
prompt-engineering-on-dtdAttriPrompt–Harmonic mean: 73.64
prompt-engineering-on-eurosatAttriPrompt–Harmonic mean: 87.12
prompt-engineering-on-fgvc-aircraftAttriPrompt–Harmonic mean: 39.80
prompt-engineering-on-imagenetAttriPrompt–Harmonic mean: 74.17
prompt-engineering-on-imagenet-aAttriPrompt–Top-1 accuracy %: 52.07
prompt-engineering-on-imagenet-rAttriPrompt–Top-1 accuracy %: 78.17
prompt-engineering-on-imagenet-v2AttriPrompt–Top-1 accuracy %: 65.07
prompt-engineering-on-oxford-102-flowerAttriPrompt–Harmonic mean: 86.62
prompt-engineering-on-oxford-iiit-pet-datasetAttriPrompt–Harmonic mean: 97.09
prompt-engineering-on-stanford-carsAttriPrompt–Harmonic mean: 77.70
prompt-engineering-on-sun397AttriPrompt–Harmonic mean: 81.10