AI Paper Atlas

Research map / Detection and segmentation / CNN features

Pixel-in-Pixel Net: Towards Efficient Facial Landmark Detection in the Wild

Haibo Jin, Shengcai Liao, Ling Shao

IJCV 2021 · Detection and segmentation › CNN features · approach: Box regression and loss design · 171 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 network stride 32 instead of 16 (ablation)PIPNet w/o NRM · WFLW validationNME (%): 4.03 -> 3.94Improved
use network stride 128 instead of 32 (ablation)PIPNet w/o NRM · WFLW validationNME (%): 3.94 -> 4.10Hurt
add neighbor regression module (NRM) (ablation)PIPNet · WFLW validationNME (%)Improved
replace standard self-training with self-training with curriculum (UDA) (ablation)PIPNet · WFLW-68NME (%): 7.45 -> 7.28Improved
replace standard self-training with self-training with curriculum (UDA) (ablation)PIPNet · COFW-68NME (%): 4.34 -> 4.28Improved
replace standard self-training with self-training with curriculum (GSSL, CelebA) (ablation)PIPNet · 300WNME (%): 3.27 -> 3.23Improved
add standard self-training with unlabeled CelebA (GSSL)PIPNet · COFW-68NME (%): 4.55 -> 4.32Improved
add DANN domain adaptation (UDA)PIPNet · 300WNME (%): 3.36 -> 3.42Hurt
add DANN domain adaptation (UDA)PIPNet · COFW-68NME (%): 4.55 -> 4.55No clear effect
use CelebA (GSSL) instead of target-domain unlabeled data (UDA)PIPNet STC · WFLW-68NME (%): 7.28 -> 7.53Hurt
use ResNet-101 instead of ResNet-18 backbonePIPNet · COFWNME (%): 3.31 -> 3.08Improved
use ResNet-101 instead of ResNet-18 backbonePIPNet · 300VW category 1NME (%): 3.04 -> 3.03No clear effect

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