AI Paper Atlas

Research map / Label-efficient and robust learning / Self-supervised and contrastive

POA: Pre-training Once for Models of All Sizes

Yingying Zhang, Xin Guo, Jiangwei Lao, Lei Yu, Lixiang Ru, Jian Wang, Guo Ye, HUIMEI HE, Jingdong Chen, Ming Yang

ECCV 2024 · Label-efficient and robust learning › Self-supervised and contrastive · approach: Invariance-based self-supervised learning · 3 citations (OpenAlex)

Read the paper

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 elastic student branch with once-for-all sub-network extraction (ViT-B from ViT-L teacher)POA · ImageNet-1K k-NN, ViT-B/16top-1 accuracy (%): 77.1 -> 80.9 (vs iBOT)Improved
add same-view distillation loss L_ES2 from intact to elastic student (ablation)POA · ImageNet-1K k-NN, ViT-S/16top-1 accuracy (%): 72.8 -> 76.2 (w/o vs with, no MPH)Improved
add cross-view elastic loss L_ES1 (ablation)POA · ImageNet-1K k-NN, ViT-S/16top-1 accuracy (%): 75.1 -> 76.2Improved
add multiple projection heads (MPH) with different prototype counts (ablation)POA · ImageNet-1K k-NN, ViT-S/16top-1 accuracy (%): 76.2 -> 76.8Improved
extract sub-networks from POA teacher instead of SEED distillation from a pre-trained ViT-L (600 epochs)ViT-S/16 · ImageNet-1K k-NNtop-1 accuracy (%): 74.0 -> 76.8Improved
drop intact student and use elastic teacher (POA-V1) (ablation)POA · ImageNet-1K k-NN, ViT-L/16top-1 accuracy (%): 82.3 -> 79.2Hurt
drop intact student, EMA teacher from elastic students (POA-V2) (ablation)POA · ImageNet-1K k-NN, ViT-L/16top-1 accuracy (%): 82.3 -> 80.9Hurt
add an extra elastic teacher branch (POA-V3) (ablation)POA · ImageNet-1K k-NN, ViT-S/16top-1 accuracy (%): 76.8 -> 76.9No clear effect
use POA-pretrained ViT-S/16 as backbone instead of iBOTCascade Mask R-CNN · COCO detectionAP^b: 49.4 -> 50.6Improved
use POA-pretrained ViT-B/16 instead of iBOT with linear segmentation headADE20K linear segmentation · ADE20KmIoU: 38.3 -> 40.3Improved
use extracted ResNet-50 vs ReLICv2 trained 4000 epochsPOA · ImageNet-1K linear probing, ResNet-50top-1 accuracy (%): 77.1 -> 76.9No clear effect
add elastic student branch to ResNet pre-trainingPOA with ResNet backbone · pre-training stabilitytraining stabilityImproved

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