GENIE: Gram-Eigenmode INR Editing with Closed-Form Geometry Updates
📰 ArXiv cs.AI
GENIE enables editing of Implicit Neural Representations (INRs) without retraining by leveraging Gram operator and deformation eigenmodes
Action Steps
- Identify the penultimate features of the INR
- Compute the Gram operator induced by these features
- Extract deformation eigenmodes from the Gram operator
- Apply closed-form geometry updates to edit the INR
Who Needs to Know This
ML researchers and engineers working with INRs can benefit from GENIE to edit and refine their models without requiring extensive retraining, making it a useful tool for applications where geometry editing is crucial
Key Insight
💡 Deformation eigenmodes parameterize a family of realizable edits of the SDF zero level set, enabling efficient geometry editing
Share This
🔍 Edit INRs without retraining with GENIE!
DeepCamp AI