Decouple before Align: Visual Disentanglement Enhances Prompt Tuning

Open paper
Benchmark Model Rank Results
prompt-engineering-on-imagenet-aDAPT + PromptKD#1Top-1 accuracy %: 69.73
prompt-engineering-on-imagenet-aDAPT + PromptSRC#5Top-1 accuracy %: 51.32
prompt-engineering-on-imagenet-aDAPT-S#16Top-1 accuracy %: 49.41
prompt-engineering-on-imagenet-aDAPT-G#19Top-1 accuracy %: 47.97
prompt-engineering-on-imagenet-rDAPT + PromptKD#1Top-1 accuracy %: 87.52
prompt-engineering-on-imagenet-rDAPT-S#14Top-1 accuracy %: 77.22
prompt-engineering-on-imagenet-rDAPT + PromptSRC#16Top-1 accuracy %: 77.14
prompt-engineering-on-imagenet-rDAPT-G#18Top-1 accuracy %: 76.82
prompt-engineering-on-imagenet-sDAPT + PromptKD#1Top-1 accuracy %: 59.07
prompt-engineering-on-imagenet-sDAPT + PromptSRC#5Top-1 accuracy %: 49.98
prompt-engineering-on-imagenet-sDAPT-S#13Top-1 accuracy %: 49.43
prompt-engineering-on-imagenet-sDAPT-G#19Top-1 accuracy %: 48.82
prompt-engineering-on-imagenet-v2DAPT + PromptKD#1Top-1 accuracy %: 70.11
prompt-engineering-on-imagenet-v2DAPT + PromptSRC#7Top-1 accuracy %: 64.81
prompt-engineering-on-imagenet-v2DAPT-G#9Top-1 accuracy %: 64.51
prompt-engineering-on-imagenet-v2DAPT-S#11Top-1 accuracy %: 64.43

Rank counts only results with a code link.