ObjectMorpher: 3D-Aware Image Editing via Deformable 3DGS Models

Dimensions overall score 7.0

• ObjectMorpher enables both 6-DoF pose and non-rigid shape manipulation within 3D space, unlike prior methods that provide only rigid pose control.method 85%
  Evidencepartial

  our approach enables manipulation of both 6-DoF pose and non-rigid shape within 3D space.

  Implicationpartial

  Directly stated in the analysis section comparing to OBJECT-3DIT and Neural Assets, which are described as providing only rigid control.

  Verificationpartial

  partial

• ObjectMorpher uses a 3D Gaussian Splatting (3DGS) representation for better fidelity and editability compared to mesh decoders.technical 90%
  Evidencepartial

  We choose the 3DGS output over the mesh decoder of TRELLIS for its better fidelity and editability... achieving higher realism and smoother gradients for manipulation.

  Implicationpartial

  Explicitly stated with reasoning about the advantages of 3DGS over mesh decoders in terms of fidelity and editability.

  Verificationpartial

  partial

• The core interactive drag stage of ObjectMorpher achieves near-instantaneous latency of 10 ms.result 90%
  Evidencepartial

  the core Interactive Drag stage achieves a near-instantaneous latency of 10 ms,

  Implicationpartial

  Direct numeric claim provided in the runtime comparison section.

  Verificationpartial

  partial

• ObjectMorpher outperforms 2D drag and 3D-aware baselines on KID, LPIPS, SIFID, and user preference metrics.result 85%
  Evidencepartial

  ObjectMorpher delivers fine-grained, photorealistic edits with superior controllability and efficiency, outperforming 2D drag and 3D-aware baselines on KID, LPIPS, SIFID, and user preference.

  Implicationpartial

  Explicitly stated in the abstract as a key result, though specific numeric values are not provided in the given excerpts.

  Verificationpartial

  partial

• Using only Laplacian deformation for mesh editing often distorts local geometry, particularly under large rotations.technical 85%
  Evidencepartial

  using only Laplacian deformation... often distorts the local geometry. This is particularly evident under large rotations.

  Implicationpartial

  Directly stated in the ablation study section comparing deformation methods.

  Verificationpartial

  partial

• ObjectMorpher's runtime is primarily spent on Object Lifting (~10 s) and Diffusion Refinement (~10 s).result 90%
  Evidencepartial

  The runtime of ObjectMorpher is primarily spent on Object Lifting (~10 s) and Diffusion Refinement (~10 s),

  Implicationpartial

  Direct numeric claim about runtime breakdown provided in the runtime comparison section.

  Verificationpartial

  partial

• Existing 3D-aware image editing methods like Image Sculpting require over half an hour per image, limiting precision and efficiency.limitation 90%
  Evidencepartial

  Image Sculpting relies on slow per-image SDS/textual inversion, limiting precision and efficiency and requiring over half an hour per image.

  Implicationpartial

  Direct numeric claim about competitor's limitations provided in the analysis section.

  Verificationpartial

  partial

• The ARAP deformation model preserves local shape rigidity while allowing intuitive, non-rigid manipulation.method 85%
  Evidencepartial

  We adopt an as-rigid-as-possible (ARAP) deformation model to preserve local shape rigidity while allowing intuitive, non-rigid manipulation.

  Implicationpartial

  Directly stated in the technical description of the deformation method.

  Verificationpartial

  partial

Competitive landscape

ObjectMorpher enables precise, 3D-aware image editing by converting 2D edits into geometry-grounded manipulations using deformable 3D Gaussian Splatting.