AI Paper Atlas

Research map / Detection and segmentation / CNN features

Improving Pixel-based MIM by Reducing Wasted Modeling Capability

Yuan Liu, Songyang Zhang, Jiacheng Chen, Zhaohui Yu, Kai Chen, Dahua Lin

ICCV 2023 · Detection and segmentation › CNN features · approach: Plain and high-resolution ViTs · 42 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
fuse a shallow layer (layer 0) with output layer for MAE reconstruction (ablation)MAE ViT-B 300 ep · ADE20KmIoU: 43.9 -> 46.4Improved
fuse a deep layer instead of a shallow layer with output layer (ablation)MAE ViT-B 300 ep · ADE20KmIoU: 43.9 -> 44.5Improved
fuse 6 layers rather than 2 (ablation)MFF_MAE · ADE20KmIoU: 46.4 -> 47.7Improved
fuse all 12 layers rather than 6 (ablation)MFF_MAE · ADE20KmIoU: 47.7 -> 46.8Hurt
use linear projection rather than none or non-linear (ablation)MFF_MAE · ADE20KmIoU: 46.2 (nonlinear) -> 47.7Improved
weighted-average fusion rather than self-attention (ablation)MFF_MAE · ImageNet-1K linear probingtop-1 accuracy: 63.0 -> 63.3No clear effect
add MFF to MAE (800 ep)MAE ViT-B · ImageNet-1K 1% low-shottop-1 accuracy: 45.4 -> 48.0Improved
add MFF to MAE (300 ep)MAE ViT-B · ImageNet-1K fine-tuningtop-1 accuracy: 82.8 -> 83.3Improved
add MFF to MAEMAE ViT-B · ImageNet-Cmean corruption error: 51.7 -> 49.0Improved
add MFF to feature-based MIM (EVA, MILAN)EVA, MILAN · ImageNet linear probingtop-1 accuracy: 69.0 -> 68.9 (EVA)No clear effect
detach shallow layers to exclude deep supervision effect (ablation)MFF_MAE · ImageNet-1K linear probingtop-1 accuracy: 67.0 -> 66.8No clear effect
add MFF to MAEMAE ViT-B 800 ep · COCO detectionAP box: 47.3 -> 48.0Improved

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