Recently, I had conversations with Shan Yang from UNC about the topic of reshaping of 3D models, actually fitting a 3D human model onto a 2D static photo. As previously met Yu Chen from metail at a graphics conference who is researching new methods on the similar topic for virtual-try. Thus, I then just felt this topic is quite promising and interesting to graphics community.

Here, the paper Parametric Reshaping of Human Bodies in Images would be a good starting point to read.

Match-and-Defom is techniqucally reasonable method. Reshape a 3D model with semantic attributes is an easier task than by low-level, local edits on a 2D image directly. Matching by pose-driven approach can resort to skeletal rigging, and deform by the concept of offset defomer. Given the correspondences between the image and model contour, body-aware image warping is then completed.

Here, as I have worked on RBF interpolation for a while. An idear came to my mind. It is, the correspondences’s displacements can be the domain of RBF for the image wapring. When reading the paper mentioned above, indeed they talked it!