Research map / Image generation / Editing and style transfer
NearID: Identity Representation Learning via Near-identity Distractors
Aleksandar Cvejic, Rameen Abdal, Abdelrahman Eldesokey, Bernard Ghanem, Peter Wonka
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 changed | Where | Measured | Effect |
|---|---|---|---|
| train MAP head with NearID loss on frozen SigLIP2 | SigLIP2 · NearID | SSR (%): 30.74 -> 99.17 | Improved |
| train MAP head with NearID loss on frozen SigLIP2 | SigLIP2 · MTG | SSR (%): 0.0 -> 35.0 | Improved |
| train MAP head with NearID loss on frozen SigLIP2 | SigLIP2 · DreamBench++ | human correlation: 0.516 -> 0.545 | Improved |
| use standard InfoNCE that ignores distractors (ablation) | SigLIP2 MAP head · NearID | SSR (%): 30.74 -> 60.97 | Mixed |
| use Circle loss with ranking (ablation) | SigLIP2 MAP head · DreamBench++ | human correlation: 0.516 -> 0.141 | Hurt |
| add oracle ranking to InfoNCE (ablation) | InfoNCE · DreamBench++ | human correlation: 0.516 -> 0.167 | Hurt |
| use L_NearID instead of InfoNCE (ablation) | SigLIP2 MAP head · DreamBench++ | human correlation: 0.555 -> 0.545 | Mixed |
| add positive cohesion term (ablation) | L_NearID · NearID / MTG | SSR (%): 99.17 -> 99.31; MTG 35.0 -> 36.0 | No clear effect |
| add NearID dataset to MTG-only training (ablation) | NearID training data · NearID | SSR (%): from 99.2% to 58.3% | Improved |
| add MTG data to NearID training (ablation) | NearID training data · NearID | SSR (%): 0.8% | Improved |
| use four inpainting engines instead of one (ablation) | NearID training data · NearID | SSR (%): 98.4% -> 86.9% | Improved |
| train on MTG only versus full pipeline (ablation) | NearID training data · MTG | SSR (%): 49.0% vs 35.0% | Mixed |
Closest papers by meaning
- Omni-ID: Holistic Identity Representation Designed for Generative Tasks (CVPR 2025 · similarity 0.86)
- Illuminating Visual Identity in Universal Multimodal Embeddings (CVPR 2026 · similarity 0.86)
- ARGUS: Stacked Multi-View Identity Mosaic Injection for Subject-Preserving Video Generation (NeurIPS 2026 · similarity 0.85)
- FaceA-Net: Facial Attribute-Driven ID Preserving Image Generation Network (AAAI 2025 · similarity 0.85)
- DreamSalon: A Staged Diffusion Framework for Preserving Identity-Context in Editable Face Generation (CVPR 2024 · similarity 0.85)
- MaSC: A Masked Similarity Metric for Evaluating Concept-Driven Generation (NeurIPS 2026 · similarity 0.84)
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.
