As-Plausible-As-Possible: Plausibility-Aware Mesh Deformation Using 2D Diffusion Priors

1KAIST, 2Adobe Research

APAP, our novel shape deformation method, enables plausibility-aware mesh deformation and preservation of fine details of the original mesh offering an interface that alters geometry by directly displacing a handle (red) along a direction (gray).


We present As-Plausible-as-Possible (APAP) mesh deformation technique that leverages 2D diffusion priors to preserve the plausibility of a mesh under user-controlled deformation. Our framework uses per-face Jacobians to represent mesh deformations, where mesh vertex coordinates are computed via a differentiable Poisson Solve. The deformed mesh is rendered, and the resulting 2D image is used in the Score Distillation Sampling (SDS) process, which enables extracting meaningful plausibility priors from a pretrained 2D diffusion model. To better preserve the identity of the edited mesh, we fine-tune our 2D diffusion model with LoRA. Gradients extracted by SDS and a user-prescribed handle displacement are then backpropagated to the per-face Jacobians, and we use iterative gradient descent to compute the final deformation that balances between the user edit and the output plausibility. We evaluate our method with 2D and 3D meshes and demonstrate qualitative and quantitative improvements when using plausibility priors over geometry-preservation or distortion-minimization priors used by previous techniques.



APAP parameterizes a triangular mesh as a per-face Jacobian field that can be updated via gradient descent. Given a textured mesh and user inputs specifying the handle(s) and anchor(s), our framework initializes a Jacobian field as a trainable parameter. During the first stage, the Jacobian field is updated via iterative optimization of $L_h$, a soft constraint that initially deforms the shape according to the user's instruction. In the following stage, the mesh is rendered using a differentiable renderer $R$ and the rendered image is provided as an input to a diffusion prior finetuned with LoRA that computes the SDS loss $L_{SDS}$. The joint optimization of $L_h$ and $L_{SDS}$ improves the visual plausibility of the mesh while conforming to the given edit instruction.

Qualitative Results from 3D Shape Deformation


Qualitative Comparisons on 2D Mesh Deformation



Please consider citing our work if you find it useful.

        title = {{As-Plausible-As-Possible: Plausibility-Aware Mesh Deformation Using 2D Diffusion Priors}},
        author = {Yoo, Seungwoo and Kim, Kunho and Kim, Vladimir G. and Sung, Minhyuk},
        booktitle = {CVPR},
        year = {2024},