Robust Prompt Tuning for Vision-Language Models with Mild Semantic Noise

Benchmark Model Rank Results
prompt-engineering-on-caltech-101ANPrompt–Harmonic mean: 96.80
prompt-engineering-on-dtdANPrompt–Harmonic mean: 73.80
prompt-engineering-on-eurosatANPrompt–Harmonic mean: 91.21
prompt-engineering-on-fgvc-aircraftANPrompt–Harmonic mean: 42.14
prompt-engineering-on-food-101ANPrompt–Harmonic mean: 91.06
prompt-engineering-on-imagenetANPrompt–Harmonic mean: 74.35
prompt-engineering-on-imagenet-aANPrompt–Top-1 accuracy %: 50.37
prompt-engineering-on-imagenet-rANPrompt–Top-1 accuracy %: 77.47
prompt-engineering-on-imagenet-sANPrompt–Top-1 accuracy %: 49.13
prompt-engineering-on-imagenet-v2ANPrompt–Top-1 accuracy %: 64.63
prompt-engineering-on-oxford-102-flowerANPrompt–Harmonic mean: 86.66
prompt-engineering-on-oxford-iiit-pet-datasetANPrompt–Harmonic mean: 96.44
prompt-engineering-on-stanford-carsANPrompt–Harmonic mean: 78.85
prompt-engineering-on-sun397ANPrompt–Harmonic mean: 81.02