AmPLe: Supporting Vision-Language Models via Adaptive-Debiased Ensemble Multi-Prompt Learning

Open paper
Benchmark Model Rank Results
prompt-engineering-on-imagenet-sMMRL + AmPLe#2Top-1 accuracy %: 50.77
prompt-engineering-on-imagenet-sMMA + AmPLe#3Top-1 accuracy %: 50.57
prompt-engineering-on-imagenet-sMaPLe + AmPLe#4Top-1 accuracy %: 50.37
prompt-engineering-on-imagenet-v2MMA + AmPLe#2Top-1 accuracy %: 65.73
prompt-engineering-on-imagenet-v2MMRL + AmPLe#3Top-1 accuracy %: 65.53
prompt-engineering-on-imagenet-v2MaPLe + AmPLe#12Top-1 accuracy %: 64.43
prompt-engineering-on-imagenet-rMMRL + AmPLe#3Top-1 accuracy %: 78.30
prompt-engineering-on-imagenet-rMMA + AmPLe#4Top-1 accuracy %: 78.07
prompt-engineering-on-imagenet-rMaPLe + AmPLe#6Top-1 accuracy %: 77.90
prompt-engineering-on-imagenet-aMaPLe + AmPLe#17Top-1 accuracy %: 48.93
prompt-engineering-on-imagenet-aMMRL + AmPLe#18Top-1 accuracy %: 48.67
prompt-engineering-on-imagenet-aMMA + AmPLe#21Top-1 accuracy %: 47.77

Rank counts only results with a code link.