AI Paper Atlas

Research map / Architectures and efficiency / Pruning and efficient tuning

Soft Masking for Cost-Constrained Channel Pruning

Ryan Humble, Maying Shen, Jorge Albericio Latorre, Eric Darve, Jose Alvarez

ECCV 2022 · Architectures and efficiency › Pruning and efficient tuning · approach: Structured network pruning · 16 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 batch normalization scaling (ablation)SMCP ResNet50 (EagleEye baseline) · ImageNet val, high pruningtop-1 accuracy and FPSImproved
add soft input channel masking instead of permanent pruning (ablation)SMCP ResNet50 (EagleEye baseline) · ImageNet val, high pruningtop-1 accuracy and FPSImproved
allow pruning entire layers (ablation)SMCP ResNet50 (EagleEye baseline) · ImageNet valtop-1 accuracy vs FPSHurt
vary target schedule (Kw,Kt,Kc) (ablation)SMCP ResNet50 70% reduction · ImageNet valtop-1 accuracy: 0.07%No clear effect
change mask recompute interval r (ablation)SMCP ResNet50 70% reduction · ImageNet valtop-1 accuracy and costHurt
solve MCK with meet-in-the-middle instead of DP (s=1000) (ablation)SMCP resource allocation · ResNet50 solve timesolve time: ~70 second -> <1 secondImproved
SMCP vs HALP at 30% latencyHALP · ImageNet val ResNet50 EagleEyetop-1 accuracy: 74.5 -> 75.1Improved
SMCP vs HALP at 80% latency (PyTorch baseline)HALP · ImageNet val ResNet50top-1 accuracy: 77.2 -> 77.1No clear effect
SMCP vs HALP at 60% latencyHALP · ImageNet val ResNet101top-1 accuracy: 78.3 -> 78.1Hurt
SMCP vs HALP on MobileNet-V1HALP · ImageNet val MobileNet-V1top-1 accuracy: 68.3 -> 68.3No clear effect
SMCP vs HALP at 30% latency FPSHALP · ImageNet val ResNet50FPS: 2755 -> 2947Improved
prune SSD512-ResNet50 with SMCPSSD512-RN50 · PASCAL VOC 07+12mAP and FPS: 2.63 point higher mAP, 12% higher FPSImproved

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