AI Paper Atlas

Research map / Architectures and efficiency / CNN design

MSPCaps: A Multi-Scale Patchify Capsule Network with Cross-Agreement Routing for Visual Recognition

Yudong Hu, Yueju Han, Rui Sun, Jinke Ren

AAAI 2026 · Architectures and efficiency › CNN design · approach: Dense matching and estimation CNNs · 0 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 multi-scale capsules (all three scales) instead of 8x8 only (ablation)MSPCaps-T · CIFAR-10accuracy (%): 87.48 -> 88.71Improved
add 8x8 coarse scale to 32x32 and 16x16 scales (ablation)MSPCaps-T · CIFAR-10accuracy (%): 81.57 -> 88.71Improved
add fine 32x32 scale to the 8x8 scale (with 16x16) (ablation)MSPCaps-T · CIFAR-10accuracy (%): 87.48 -> 88.71Improved
use 2 cross-agreement routing blocks instead of 4 dynamic routing blocks (ablation)MSPCaps-T · CIFAR-10accuracy (%): 87.46 -> 88.71Improved
use 2 cross-agreement routing blocks instead of 4 dynamic routing blocks (ablation)MSPCaps-L · CIFAR-10accuracy (%): 92.56 -> 92.88Improved
share transformation weights within CAR (ablation)MSPCaps-T · CIFAR-10accuracy (%): 87.59 -> 88.71Improved
share transformation weights within CAR (ablation)MSPCaps-L · CIFAR-10accuracy (%): 92.88 -> 92.50Hurt
patch size 4 instead of 3 (ablation)MSPCaps-T · CIFAR-10accuracy (%): 86.34 -> 88.71Improved
patch size 4 instead of 2 (ablation)MSPCaps-T · CIFAR-10accuracy (%): 88.01 -> 88.71Improved
patch size 3 instead of 4 (ablation)MSPCaps-L · CIFAR-10accuracy (%): 92.88 -> 93.30Improved
patch size 2 instead of 4 (ablation)MSPCaps-L · CIFAR-10accuracy (%): 92.88 -> 93.16 (10.9M -> 39.4M params)Mixed
smaller model (MSPCaps-T) under adversarial attack versus CapsNetMSPCaps-T · CIFAR-10 FGSM/BIMaccuracyHurt

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