AI Paper Atlas

Research map / Detection and segmentation / CNN features

Differentiable Kernel Evolution

Yu Liu, Jihao Liu, Ailing Zeng, Xiaogang Wang

ICCV 2019 · Detection and segmentation › CNN features · approach: Efficient DETR query design · 5 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 conv layers with DKEAlexNet · ImageNettop-1 error (%): 42.8 -> 39.1Improved
replace conv layers with DKEResNet-18 · ImageNettop-1 error (%): 30.8 -> 27.15Improved
replace conv layers with DKEResNet-101 · ImageNettop-1 error (%): 22.7 -> 21.1Improved
replace conv layers with DKEAlexNet · CIFAR-100top-1 error (%): 56.30 -> 53.1Improved
replace conv layers with DKEResNet-101 · CIFAR-10top-1 error (%): 5.3 -> 4.7Improved
replace DCN by DKE layersResNet-18 · MegaFace 1 vs 1 milliontop-1 accuracy (%): 69.5 -> 91.5Improved
replace conv layers with DKEResNet-18 · MegaFace 1 vs 1 milliontop-1 accuracy (%): 89.4 -> 91.5Improved
replace conv layers with DKEResNet-101 · MegaFace 1 vs 1 milliontop-1 accuracy (%): 97.8 -> 98.0Improved
replace conv layers with deformable convolution (DCN)ResNet-18 · MegaFace 1 vs 1 milliontop-1 accuracy (%): 89.4 -> 69.5Hurt
replace conv layers with DKEFaster-RCNN Res18 · COCO 2017 miniValmAP[.5:.95]: 28.0 -> 29.6Improved
replace conv layers with DKEFPN Res50 · COCO 2017 miniValmAP[.5:.95]: 36.3 -> 36.7Improved
enlarge kernel to 5x5 instead of 3x3 (ablation)ResNet-6 · ImageNettop-1 error (%): 51.7 -> 46.6Improved

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