AI Paper Atlas

Research map / Detection and segmentation / CNN features

Evolved Hierarchical Masking for Self-Supervised Learning

Zhanzhou Feng, Shiliang Steven Zhang

TPAMI 2024 · Detection and segmentation › CNN features · approach: Convolution-transformer backbones · 9 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
evolve masking depth during training instead of fixed shallow depth (ablation)evolved hierarchical masking · ImageNet-1K top-1 (ViT-S)top-1 accuracy: 79.05 -> 80.41Improved
evolve masking depth instead of fixed shallow depth (ablation)evolved hierarchical masking · ADE20K (ViT-S)mIoU: 39.86 -> 41.37Improved
evolve masking depth instead of fixed deep depth (ablation)evolved hierarchical masking · Oxford retrieval (ViT-S)mAP: 55.07 -> 57.41Improved
evolve masking depth instead of fixed shallow depth (ablation)evolved hierarchical masking · Oxford retrieval (ViT-S)mAP: 57.11 -> 57.41No clear effect
use hierarchical shallow masking instead of grid-like masking (ablation)hierarchical masking · ImageNet-1K top-1 (ViT-S)top-1 accuracy: 73.16 -> 79.05Improved
use hierarchical deep masking instead of large block masking (ablation)hierarchical masking · ImageNet-1K top-1 (ViT-S)top-1 accuracy: 77.89 -> 78.76Improved
use hierarchical masking instead of random masking at mask ratio 0.25 (ablation)MAE · ImageNet-1K top-1 (ViT-S)top-1 accuracy: 77.19 -> 78.96Improved
use hierarchical masking instead of random masking at mask ratio 0.75 (ablation)MAE · ADE20K (ViT-S)mIoU: 39.24 -> 40.59Improved
add evolved hierarchical maskingMAE · ImageNet-1K top-1 (200 ep pretrain)top-1 accuracy: 79.20 -> 80.06Improved
add evolved hierarchical maskingSimMIM · ImageNet-1K top-1 (200 ep pretrain)top-1 accuracy: 78.51 -> 79.21Improved
add evolved hierarchical maskingMAE · ADE20K (200 ep pretrain)mIoU: 38.81 -> 40.23Improved
extend pretraining from 100 to 2400 epochsevolved hierarchical masking · ImageNet-1K top-1 (ViT-S)top-1 accuracy: 78.29 -> 81.55Improved

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