AI Paper Atlas

Research map / Image generation / Editing and style transfer

Face Image Retrieval With Attribute Manipulation

Alireza Zaeemzadeh, Shabnam Ghadar, Baldo Faieta, Zhe Lin, Nazanin Rahnavard, Mubarak Shah, Ratheesh Kalarot

ICCV 2021 · Image generation › Editing and style transfer · approach: Identity-preserving face manipulation · 19 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 L1 sparsity (lambda 1e-3) to orthogonal attribute directions (ablation)Ours · CelebA sparsity ablation (top-5, 1000 queries)nDCG: 0.826 -> 0.847Improved
raise sparsity lambda from 1e-3 to 5e-3 (ablation)Ours · CelebA sparsity ablation (top-5, 1000 queries)nDCG: 0.847 -> 0.858Improved
raise sparsity lambda from 5e-3 to 1e-2 (ablation)Ours · CelebA sparsity ablation (top-5, 1000 queries)nDCG: 0.858 -> 0.849Hurt
add sparsity (lambda 5e-3) on identity similarity (ablation)Ours · CelebA sparsity ablation (top-5, 1000 queries)identity similarity: 0.863 -> 0.864No clear effect
enforce orthogonality only (lambda=0) vs InterFaceGAN (ablation)InterFaceGAN · CelebA sparsity ablation (top-5, 1000 queries)identity similarity: 0.859 -> 0.863Improved
learn sparse orthogonal directions vs InterFaceGAN, no preferenceInterFaceGAN · CelebA attribute-guided retrieval (1000 queries, single attribute) top-5nDCG: 0.568 -> 0.595Improved
learn sparse orthogonal directions vs InterFaceGAN, identity constrainedInterFaceGAN · CelebA attribute-guided retrieval (1000 queries, single attribute) top-5nDCG: 0.822 -> 0.858Improved
learn sparse orthogonal directions vs InterFaceGAN, best nDCGInterFaceGAN · CelebA attribute-guided retrieval (1000 queries, single attribute) top-5nDCG: 0.905 -> 0.923Improved
assign per-query attribute preference (Best nDCG) instead of no preference (ablation)Ours · CelebA attribute-guided retrieval (1000 queries, single attribute) top-5nDCG: 0.595 -> 0.923Improved
increase preference for the target attribute (ablation)Ours · CelebA attribute-guided retrieval (1000 queries, single attribute) top-5identity similarity: 0.84 -> 0.82Hurt
GAN latent method with no preference vs compositional learning TIRG++TIRG++ · CelebA attribute-guided retrieval (1000 queries, single attribute) top-5nDCG: 0.822 -> 0.595Hurt

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