AI Paper Atlas

Research map / Label-efficient and robust learning / Semi-supervised learning

Aggregation via Separation: Boosting Facial Landmark Detector With Semi-Supervised Style Translation

Shengju Qian, Keqiang Sun, Wayne Wu, Chen Qian, Jiaya Jia

ICCV 2019 · Label-efficient and robust learning › Semi-supervised learning · approach: GAN-based semi-supervised learning · 101 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 style-translation augmentationSAN · WFLW fullsetNME (%): 5.22 -> 4.39Improved
add style-translation augmentationLAB · WFLW fullsetNME (%): 5.27 -> 4.76Improved
add style-translation augmentationRes-18 · WFLW fullsetNME (%): 6.09 -> 5.25Improved
add style-translation augmentationSAN · AFLW-68 full setNME (%): 6.94 -> 6.01Improved
add style-translation augmentationRes-18 · AFLW-68 full setNME (%): 7.23 -> 7.17No clear effect
add style-translation augmentation on 10% of training dataRes-18 · 300WNME (%): 13.72 -> 7.86Improved
add style-translation augmentation on 10% of training dataSAN · WFLWNME (%): 9.16 -> 7.2Improved
increase number of random styles k from 0 to 8 (ablation)Ours with ResNet-50 · WFLW test setNME (%): 6.22 -> 5.31Improved
increase number of random styles k from 16 to 32 (ablation)Ours with ResNet-50 · WFLW test setNME (%): 5.29 -> 5.34Hurt
add KL divergence loss to the disentangling model (ablation)Ours · WFLW test setNME (%): 9.08 -> 7.98Improved
add perceptual loss to the disentangling model (ablation)Ours · WFLW test setNME (%): 8.34 -> 7.98Improved
replace Res-18 with SAN backbone under style augmentationOurs · WFLW fullsetNME (%): 5.25 -> 4.39Improved

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