AI Paper Atlas

Research map / Image generation / Editing and style transfer

DVG-Face: Dual Variational Generation for Heterogeneous Face Recognition

Chaoyou Fu, Xiang Wu, Yibo Hu, Huaibo Huang, Ran He

TPAMI 2021 · Image generation › Editing and style transfer · approach: Diffusion face generation and editing · 144 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 identity sampling from large-scale VIS data and pairwise identity loss (DVG-Face vs DVG)DVG · Tufts FaceRank-1 (%): 56.1 -> 75.7Improved
add angular orthogonal loss (ablation)DVG-Face · Tufts FaceRank-1 (%): 55.9 -> 75.7Improved
add pairwise identity preserving loss (ablation)DVG-Face · Tufts FaceRank-1 (%): 62.4 -> 75.7Improved
use identity sampler in both training and testing instead of LightCNN (ablation)DVG-Face · Tufts FaceRank-1 (%): 74.6 -> 75.7Improved
use generated data in addition to real data (ablation)HFR network · Tufts FaceRank-1 (%): 54.5 -> 75.7Improved
train on generated data only without real labeled data (ablation)HFR network · Tufts FaceRank-1 (%): 54.5 -> 36.6Hurt
use contrastive loss instead of pairwise distance loss on generated data (ablation)DVG-Face · Tufts FaceRank-1 (%): 70.3 -> 75.7Improved
introduce MS-Celeb-1M identities into generation instead of fine-tuning on MS-Celeb-1M images (ablation)LightCNN · Tufts FaceRank-1 (%): 33.9 -> 75.7Improved
increase generated pairs from 0 to 100K (ablation)DVG-Face · Tufts FaceVR@FAR=1% (%): improved by 26.1%Improved
increase generated pairs beyond 100K (ablation)DVG-Face · Tufts Facerecognition performanceNo clear effect
set lambda1 to 25 instead of 0 (ablation)DVG-Face · Tufts FaceVR@FAR=1% (%): 43.6 -> 68.5Improved
set eta to 0.10 instead of 0 (ablation)DVG-Face · Tufts FaceVR@FAR=1% (%): 46.5 -> 68.5Improved

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