AI Paper Atlas

Research map / Architectures and efficiency / CNN design

Convolutional Networks with Adaptive Inference Graphs

Andreas Veit, Serge Belongie

ECCV 2018 · Architectures and efficiency › CNN design · approach: Efficient CNN scaling and dynamic inference · 373 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
replace ResNet 110 with ConvNet-AIG 110 (gated, executes subset of layers)ResNet 110 · CIFAR-10test error (%): 6.61 -> 5.76Improved
execute all layers with outputs scaled by expected execution rate (ConvNet-AIG 110*) (ablation)ConvNet-AIG 110 · CIFAR-10test error (%): 5.76 -> 5.14Improved
replace ResNet 50 (ours) with ConvNet-AIG 50 at target rate 0.7ResNet 50 · ImageNet valtop-1 error (%): 23.87 -> 23.82Improved
replace ResNet 101 (ours) with ConvNet-AIG 101 at target rate 0.5ResNet 101 · ImageNet valtop-1 error (%): 22.63 -> 22.63No clear effect
lower target rate for ConvNet-AIG 50 from 0.7 to 0.4 (ablation)ConvNet-AIG 50 · ImageNet valtop-1 error (%): 23.82 -> 24.75Hurt
lower target rate for ConvNet-AIG 50 from 0.7 to 0.4 (ablation)ConvNet-AIG 50 · ImageNet valFLOPs (x10^9): 3.06 -> 2.56Improved
apply stochastic depth to ResNet 50ResNet 50 · ImageNet valtop-1 error (%): 23.87 -> 27.75Hurt
apply stochastic depth to ResNet 101ResNet 101 · ImageNet valtop-1 error (%): 22.63 -> 22.80Improved
use ConvNet-AIG gating instead of plain ResNet under FGSM adversarial attackResNet · ImageNet val (FGSM)top-1 accuracyImproved
always execute early layers in ConvNet-AIG 101 (ablation)ConvNet-AIG 101 · ImageNet valdesign choiceNo 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