AI Paper Atlas

Research map / Vision-language and multimodal / CLIP and zero-shot

Benchmarking Robustness of Adaptation Methods on Pre-trained Vision-Language Models

Shuo Chen, Jindong Gu, Zhen Han, Yunpu Ma, Philip Torr, Volker Tresp

NeurIPS 2023 · Vision-language and multimodal › CLIP and zero-shot · approach: Test-time adaptation of VLMs · 7 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
replace full fine-tuning with a single adapterCLIP-BART · VQAv2 text corruptionsRR (%): 73.65 -> 77.64Improved
replace full fine-tuning with a single adapterCLIP-BART · VQAv2 image corruptionsRR (%): 84.86 -> 85.76Improved
replace full fine-tuning with a single adapterCLIP-BART · GQA image corruptionsRR (%): 89.20 -> 82.49Hurt
replace full fine-tuning with a single adapterCLIP-BART · GQA cleanaccuracy (%): 55.04 -> 54.14Hurt
use half-shared adapters instead of a single adapterCLIP-BART · GQA image corruptionsRR (%): 82.49 -> 89.37Improved
use multiple adapters instead of a single adapterCLIP-BART · GQA image corruptionsRR (%): 82.49 -> 86.16Improved
use multiple LoRA instead of single LoRACLIP-BART · GQA image corruptionsRR (%): 82.58 -> 91.15Improved
use multiple LoRA instead of single LoRACLIP-BART · GQA text corruptionsRR (%): 63.94 -> 68.77Improved
use Hyperformer instead of full fine-tuningCLIP-BART · GQA text corruptionsRR (%): 66.92 -> 70.30Improved
apply text corruptions instead of image corruptionsSingle Adapter on CLIP-BART · VQAv2RR (%): 85.76 -> 77.64Hurt
increase adaptation data from 20% to 100% (ablation)adaptation methods · all tasksrelative robustnessHurt
increase soft prompt length (ablation)prompt tuning on CLIP-BART · four tasksclean performanceImproved

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