AI Paper Atlas

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

MemBN: Robust Test-Time Adaptation via Batch Norm with Statistics Memory

Juwon Kang, Nayeong Kim, Jungseul Ok, Suha Kwak

ECCV 2024 · Label-efficient and robust learning › Test-time adaptation · approach: Realistic and stable test-time adaptation · 8 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 statistics memory of size 128 with reset (vs TBN) (ablation)TBN · CIFAR10-C continualavg error rate (%): 33.09 -> 15.07Improved
add memory reset to a 128-size memory (ablation)MemBN memory · CIFAR10-C continualavg error rate (%): 17.40 -> 15.07Improved
add adaptive weighting between source and memory statistics (ablation)MemBN · CIFAR10-C continualavg error rate (%): 15.07 -> 12.18Improved
add adaptive weighting between source and memory statistics (ablation)MemBN · CIFAR100-C continualavg error rate (%): 40.17 -> 36.80Improved
use memory of 128 without reset instead of EMA memory of size 1 (ablation)MemBN · CIFAR10-C continual, batch size 200error rate (%): 14.63 -> 21.05Hurt
replace in-batch statistics with MemBN statisticsTENT · ImageNet-C batch size 1error rate (%): 99.90 -> 67.88Improved
add MemBN to EATAEATA · ImageNet-Cavg error rate (%): 77.15 -> 58.42Improved
add MemBN to EATAEATA · ImageNet-C batch size 1error rate (%): 99.86 -> 73.72No clear effect
use MemBN instead of TTN without any post-training phaseTTN · CIFAR10-C single domainavg error rate (%): 13.89 -> 12.48Improved
use MemBN at large batch size 200 vs TTNTTN · CIFAR10-C single domain, batch size 200error rate (%): 11.67 -> 12.68Hurt
add MemBN to TENT in segmentationTENT · Cityscapes to BDD-100K, batch size 1mIoU: 40.78 -> 50.94Improved

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 3 of 3 on most shared results.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

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