AI Paper Atlas

Research map / Detection and segmentation / CNN features

Lightweight Pixel Difference Networks for Efficient Visual Representation Learning

Zhuo Su, Jiehua Zhang, Longguang Wang, Hua Zhang, Zhen Liu, Matti Pietikäinen, Li Liu

TPAMI 2023 · Detection and segmentation › CNN features · approach: Convolution-transformer backbones · 70 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 vanilla convolution with PDC in all 16 blocks (ablation)PiDiNet · BSDS500 valODS: 0.772 -> 0.767Hurt
use sequence of four conv types [CARV]x4 instead of single type repeated (ablation)PiDiNet · BSDS500 valODS: 0.772 -> 0.776Improved
add PDC in only the first block (ablation)PiDiNet · BSDS500 valODS: 0.772 -> 0.775Improved
add PDC at multiple scales vs vanilla convolution (ablation)PiDiNet · BSDS500 val (Basic C=60)ODS: 0.736 -> 0.757Improved
add shortcuts (ablation)PiDiNet · BSDS500 valODS: 0.734 -> 0.776Improved
add CSAM (ablation)PiDiNet · BSDS500 valODS: 0.775 -> 0.776Improved
add CDCM and CSAM (ablation)PiDiNet-L · BSDS500 valODS: 0.770 -> 0.776Improved
replace vanilla conv with PDCBaseline · BSDS500 testODS: 0.798 -> 0.807Improved
replace vanilla conv with PDCBaseline · BSDS500 testGPU FPS: 96 -> 92Hurt
fix binary kernels as LBP patterns instead of learning them (ablation)Bi-PDC · CIFAR-100 (binary ResNet-18)top-1 accuracy: 70.49 -> 68.80Hurt
add Bi-PDC at xi=0.2 channel ratio (ablation)binary ResNet-18 · CIFAR-100top-1 accuracy: performance rises up to xi 0.2 then dropsMixed
replace vanilla binary conv with Bi-CPDC (ablation)binary ResNet-18 · ImageNettop-1 accuracy: 60.3 -> 61.6Improved

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 0 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