AI Paper Atlas

Research map / Label-efficient and robust learning / Test-time adaptation

SwapPrompt: Test-Time Prompt Adaptation for Vision-Language Models

Xiaosong Ma, Jie ZHANG, Song Guo, Wenchao Xu

NeurIPS 2023 · Label-efficient and robust learning › Test-time adaptation · approach: Pseudo-label test-time adaptation · 14 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
add swapped prediction loss L_swap on top of augmented pseudo-label loss (ablation)SwapPrompt · 6 datasets avgaccuracy: 69.39 -> 70.68Improved
add image augmentation to UPL (ablation)UPL · 6 datasetsaccuracy: 68.65 -> 69.39Improved
use EMA target prompt (0.99) instead of fixed target prompt (ablation)SwapPrompt · 6 datasets avgaccuracy: 69.33 -> 70.68Improved
use EMA target prompt instead of target identical to online prompt (decay 0) (ablation)SwapPrompt · 6 datasets avgaccuracy: 69.65 -> 70.68Improved
swapped prediction L_swap versus L1 (no prompt interaction) (ablation)SwapPrompt · 5 datasets avgaccuracy: 71.32 -> 72.45Improved
swapped prediction L_swap versus L2 (no cross-view contrast) (ablation)SwapPrompt · 5 datasets avgaccuracy: 71.24 -> 72.45Improved
add top-K confident data selection (K=16) (ablation)SwapPrompt · 5 datasets avgaccuracy: 69.95 -> 72.45Improved
increase K from 1 to 16 (ablation)SwapPrompt · 5 datasets avgaccuracy: 68.74 -> 72.45Improved
increase context length from 4 to 16 tokens (ablation)SwapPrompt · 5 datasets avgaccuracy: 72.45 -> 71.83Hurt
hand-crafted or pre-ImageNet prompt initialization instead of random (ablation)SwapPrompt · 5 datasets avgaccuracy: 72.45 vs 72.33 / 72.36No clear effect
change alpha/beta weights from 1/1 (ablation)SwapPrompt · 5 datasets avgaccuracy: 72.45 -> 71.77 / 72.12No clear effect
online mini-batch test-time adaptation instead of full offlineSwapPrompt · EuroSATaccuracy: 46.64 -> 42.45Mixed

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy