AI Paper Atlas

Research map / Architectures and efficiency / CNN design

Factorized Bilinear Models for Image Recognition

Yanghao Li, Naiyan Wang, Jiaying Liu, Xiaodi Hou

ICCV 2017 · Architectures and efficiency › CNN design · approach: Filter design and trained-filter analysis · 114 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 a factorized bilinear layer before global poolingInception-BN · CIFAR-100test error (%): 24.70 -> 21.98Improved
add FB layer in the lower In5a factory (ablation)Inception-BN · CIFAR-100test error (%): 24.70 -> 24.73No clear effect
apply FB per spatial position (Conv-FBN) instead of in the final FC layer (ablation)Inception-BN · CIFAR-100test error (%): 24.07 -> 21.98Improved
add a second FB block (In5b) on top of Conv-FBN (ablation)Inception-BN · CIFAR-100test error (%): 21.98 -> 23.70Hurt
increase number of factors k from 10 to 20 (ablation)Conv-FBN · CIFAR-100test error (%): 22.99 -> 21.98Improved
increase number of factors k from 20 to 80 (ablation)Conv-FBN · CIFAR-100test error (%): 21.98 -> 21.88No clear effect
add DropFactor (ablation)FBN (Inception-BN) · CIFAR-100test error (%): 23.55 -> 21.98Improved
add Dropout on top of DropFactor (ablation)FBN (Inception-BN) · CIFAR-100test error (%): 21.98 -> 22.71Hurt
use 3x3 instead of 1x1 FB kernel (ablation)Conv-FBN · CIFAR-100test error (%): 21.98 -> 23.08Hurt
add FB layer (Conv-FBN)ResNet-1001 · CIFAR-100top-1 error (%): 20.50 -> 19.67Improved
add FB layer (Conv-FBN)ResNet-1001 · CIFAR-10top-1 error (%): 4.04 -> 4.09No clear effect
add FB layer (Conv-FBN)ResNet-50 · ImageNet valtop-1 error (%): 24.7 -> 24.0Improved

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