AI Paper Atlas

Research map / Detection and segmentation / CNN features

Multiscale Deep Equilibrium Models

Shaojie Bai, Vladlen Koltun, J. Zico Kolter

NeurIPS 2020 · Detection and segmentation › CNN features · approach: Efficient network training and design · 75 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
use simultaneous multiscale equilibrium features instead of a single-stream DEQMDEQ-small · CIFAR-10 without data augmentationaccuracy: 82.2% ± 0.3% -> 87.1% ± 0.4%Improved
use simultaneous multiscale equilibrium features instead of a single-stream implicit modelMDEQ-small · ImageNet classificationtop-1 accuracy: 72.9% -> 75.5%Improved
use MDEQ instead of ResNet-18 at similar model sizeCIFAR-10 without data augmentation · CIFAR-10 without data augmentationaccuracy: 81.6% ± 0.3% -> 87.1% ± 0.4%Improved
use MDEQ instead of ResNet-18 at similar model sizeCIFAR-10 with data augmentation · CIFAR-10 with data augmentationaccuracy: 92.9% ± 0.2% -> 93.8% ± 0.3%Improved
use equilibrium solving instead of a 5-layer unrolled MDEQ (ablation)MDEQ-large · ImageNet classificationtop-1 accuracy: 75.9% -> 77.5%Improved
use MDEQ instead of ResNet-50MDEQ-small · ImageNet classificationtop-1 accuracy: 75.1% -> 75.5%Improved
use MDEQ instead of MobileNetV2PlusMDEQ-small · Cityscapes val semantic segmentationmIoU: 74.5 -> 75.1Improved
use MDEQ instead of DeepLabv3MDEQ-XL · Cityscapes val semantic segmentationmIoU: 78.5 -> 80.3Improved
use Broyden's method instead of repeatedly iterating the transformationMDEQ · CIFAR-10, ImageNet, and Cityscapes equilibrium solvingequilibrium convergence stabilityImproved
use simultaneous multiscale fusion instead of a single-stream equilibriumMDEQ · CIFAR-10 equilibrium solvinghigh-resolution stream convergenceImproved

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

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