AI Paper Atlas

Research map / Architectures and efficiency / CNN design

Latency-Aware Unified Dynamic Networks for Efficient Image Recognition

Yizeng Han, Zeyu Liu, Zhihang Yuan, Yifan Pu, Chaofei Wang, Shiji Song, Gao Ming Huang

TPAMI 2024 · Architectures and efficiency › CNN design · approach: Efficient CNN scaling and dynamic inference · 34 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 masker with the first convolution (ablation)LAUD^s-ResNet-50 block · Operator fusion, V100latency (us): 162.4 -> 135.1Improved
also fuse gather with convolution (ablation)LAUD^s-ResNet-50 block · Operator fusion, V100latency (us): 135.1 -> 131.7Improved
also fuse scatter with add (ablation)LAUD^s-ResNet-50 block · Operator fusion, V100latency (us): 131.7 -> 118.3Improved
add full operator fusion (ablation)LAUD^s-ResNet-50 block · Operator fusion, TX2latency (us): 1084.5 -> 859.7Improved
use coarse spatial granularity (S>1) instead of pixel-level (S=1) (ablation)spatially dynamic convolution · latency ratio on V100 and TX2latency ratio r_l: S=1 does not consistently give speedup; S>1 attains r_l<1 at larger activation ratesImproved
increase S from 7 to 14 in second stage (ablation)LAUD-RegNetY-800MF · TX2latency: detrimentally impacts efficiencyHurt
raise channel granularity G from 1 to 2 (ablation)LAUD^c-ResNet-50 · ImageNet latencylatency: speedup for ResNet-50Improved
raise channel granularity G from 1 to 2 (ablation)LAUD^c-RegNetY-800M · ImageNet latencylatency: comparable performanceNo clear effect
use layer skipping instead of spatial-wise dynamic convolutionLAUD-ResNet-101 · V100 ImageNet latencylatency: ~53% latency reduction at t=0.4Improved
use layer skipping instead of channel skipping (ablation)LAUD-T2T-ViT · TX2, RTX3060, V100accuracy-latency trade-off: consistently outperforms head skippingImproved
combine all three paradigms (ablation)LAUD-T2T-ViT · ImageNetaccuracy-efficiency trade-off: further enhances trade-offImproved
use LAUD^s-ResNet-101 (t=0.5) backboneMask R-CNN · COCO instance segmentationAP mask: 36.1 -> 37.0Improved

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