AI Paper Atlas

Research map / Detection and segmentation / CNN features

Latency-aware Spatial-wise Dynamic Networks

Yizeng Han, Zhihang Yuan, Yifan Pu, Chenhao Xue, Shiji Song, Guangyu Sun, Gao Huang

NeurIPS 2022 · Detection and segmentation › CNN features · approach: Convolution-transformer backbones · 11 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 and first conv (ablation)LAS-ResNet-101 block · stage-1 block, V100latency (us): 163.2 -> 90.1Improved
fuse gather with dynamic conv (ablation)LAS-ResNet-101 block · stage-1 block, V100latency (us): 90.1 -> 86.7Improved
fuse scatter with add (ablation)LAS-ResNet-101 block · stage-1 block, V100latency (us): 86.7 -> 71.4Improved
use coarse spatial granularity S>1 instead of pixel-level S=1 (ablation)LAS-ResNet-101 · ImageNet, V100latencyImproved
use coarse granularity (4-4-2-1, 8-4-7-1) vs 1-1-1-1 (ablation)LAS-ResNet · ImageNetaccuracy at similar FLOPsImproved
use LAS-ResNet-101 instead of ResNet-101 at t=0.4ResNet-101 · ImageNetlatency: -36% on V100, -46% on TX2Improved
replace S=4-4-2-1 with layer skipping S=56-28-14-7 (ablation)LAS-ResNet-50 · ImageNetaccuracy (%): 76.6 -> 76.1Hurt
use LAS-ResNet-101 (t=0.6) as Faster R-CNN backboneResNet-101 · COCO Faster R-CNNmAP (%): 39.4 -> 40.3Improved
use LAS-ResNet-101 (t=0.4) as Faster R-CNN backboneResNet-101 · COCO Faster R-CNNbackbone latency V100 (ms): 39.5 -> 25.3Improved
use LAS-ResNet-101 (t=0.4) as RetinaNet backboneResNet-101 · COCO RetinaNetmAP (%): 38.5 -> 38.9No clear effect
use LAS-ResNet-101 (t=0.4, 4-4-2-1) in Mask R-CNNResNet-101 · COCO Mask R-CNNAP mask (%): 36.1 -> 36.1No clear effect

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