Gram-Anchored Prompt Learning for Vision-Language Models via Second-Order Statistics

Benchmark Model Rank Results
prompt-engineering-on-caltech-101GAPL–Harmonic mean: 97.36
prompt-engineering-on-dtdGAPL–Harmonic mean: 76.13
prompt-engineering-on-eurosatGAPL–Harmonic mean: 87.05
prompt-engineering-on-fgvc-aircraftGAPL–Harmonic mean: 44.31
prompt-engineering-on-food-101GAPL–Harmonic mean: 91.48
prompt-engineering-on-imagenetGAPL–Harmonic mean: 73.40
prompt-engineering-on-imagenet-aGAPL–Top-1 accuracy %: 50.43
prompt-engineering-on-imagenet-rGAPL–Top-1 accuracy %: 77.50
prompt-engineering-on-imagenet-sGAPL–Top-1 accuracy %: 50.13
prompt-engineering-on-imagenet-v2GAPL–Top-1 accuracy %: 66.43
prompt-engineering-on-oxford-102-flowerGAPL–Harmonic mean: 86.92
prompt-engineering-on-oxford-iiit-pet-datasetGAPL–Harmonic mean: 96.80
prompt-engineering-on-stanford-carsGAPL–Harmonic mean: 79.27
prompt-engineering-on-sun397GAPL–Harmonic mean: 81.81
prompt-engineering-on-ucf101GAPL–Harmonic mean: 83.34