TriAnchor-ID: Multi-View Semantic Identity Anchoring for Pose-Aware Personalized Diffusion
Fuente:
Zenodo
Enregistré dans:
| Auteur principal: | |
|---|---|
| Format: | Recurso digital |
| Langue: | anglais |
| Publié: |
Zenodo
2026
|
| Sujets: | |
| Accès en ligne: | |
| Tags: |
Ajouter un tag
Pas de tags, Soyez le premier à ajouter un tag!
|
| _version_ | 1866901272183963648 |
|---|---|
| author | KUMAR, AKASH |
| author_facet | KUMAR, AKASH |
| contents | <p>This paper presents TriAnchor-ID, a prototype framework for identity-conditioned diffusion portrait generation. The work focuses on multi-view semantic identity anchoring using ArcFace/InsightFace embeddings, quality-weighted identity aggregation, landmark-derived geometric observations, and post-generation identity consistency evaluation.</p> <p>The paper corrects an important distinction between semantic face identity embeddings and physical 3D morphable model parameters: the implemented 512-dimensional ArcFace vector is treated as a semantic identity anchor, not as a FLAME or 3DMM shape vector. The proposed Identity Capsule design separates implemented components from future extensions such as FLAME/DECA shape fitting, UV texture modeling, spatial ControlNet conditioning, and inference-time identity consistency guidance.</p> <p>This version is released as a preprint/prototype research report. The qualitative examples are illustrative, and the paper outlines a controlled evaluation protocol using ArcFace similarity, identity drift score, landmark error, face detection failure rate, prompt alignment, repeated seeds, and confidence intervals.</p> |
| format | Recurso digital |
| id | zenodo_https___doi_org_10_5281_zenodo_19836585 |
| institution | Zenodo |
| language | eng |
| publishDate | 2026 |
| publisher | Zenodo |
| record_format | zenodo |
| spellingShingle | TriAnchor-ID: Multi-View Semantic Identity Anchoring for Pose-Aware Personalized Diffusion KUMAR, AKASH personalized diffusion identity-conditioned generation ArcFace IP-Adapter FaceID computer vision generative AI diffusion models <p>This paper presents TriAnchor-ID, a prototype framework for identity-conditioned diffusion portrait generation. The work focuses on multi-view semantic identity anchoring using ArcFace/InsightFace embeddings, quality-weighted identity aggregation, landmark-derived geometric observations, and post-generation identity consistency evaluation.</p> <p>The paper corrects an important distinction between semantic face identity embeddings and physical 3D morphable model parameters: the implemented 512-dimensional ArcFace vector is treated as a semantic identity anchor, not as a FLAME or 3DMM shape vector. The proposed Identity Capsule design separates implemented components from future extensions such as FLAME/DECA shape fitting, UV texture modeling, spatial ControlNet conditioning, and inference-time identity consistency guidance.</p> <p>This version is released as a preprint/prototype research report. The qualitative examples are illustrative, and the paper outlines a controlled evaluation protocol using ArcFace similarity, identity drift score, landmark error, face detection failure rate, prompt alignment, repeated seeds, and confidence intervals.</p> |
| title | TriAnchor-ID: Multi-View Semantic Identity Anchoring for Pose-Aware Personalized Diffusion |
| topic | personalized diffusion identity-conditioned generation ArcFace IP-Adapter FaceID computer vision generative AI diffusion models |
| url | https://doi.org/10.5281/zenodo.19836585 |