Register now After registration you will be able to apply for this opportunity online.
This opportunity is not published. No applications will be accepted.
Recovering Blurry Human Body
When humans move, the images are usually corrupted by motion blur. All previous methods for human body and shape reconstruction assume sharp images. In this thesis, the goal is to design a method that could deal with motion blur.
Keywords: human body reconstruction, motion blur, 3d reconstruction
As a starting point, the method can use Shape from Blur [1] codebase. Instead of optimizing the prototypes and their deformations, an SMPL body model [2] should be used. This low-space latent representation of the human body will become part of optimization, with initial body motion and final body motion (and linear interpolation in-between). For qualitative evaluation, the student should gather a dataset with blurry human bodies. For quantitative evaluation, some videos with sharp human bodies can be used to create a blurry input by averaging frames.
[1] Rozumnyi et al. "Shape from Blur: Recovering Textured 3D Shape and Motion of Fast Moving Objects", NeurIPS 2021, Code: https://github.com/rozumden/ShapeFromBlur
[2] Loper et al, "SMPL: A Skinned Multi-Person Linear Model"
As a starting point, the method can use Shape from Blur [1] codebase. Instead of optimizing the prototypes and their deformations, an SMPL body model [2] should be used. This low-space latent representation of the human body will become part of optimization, with initial body motion and final body motion (and linear interpolation in-between). For qualitative evaluation, the student should gather a dataset with blurry human bodies. For quantitative evaluation, some videos with sharp human bodies can be used to create a blurry input by averaging frames.
[1] Rozumnyi et al. "Shape from Blur: Recovering Textured 3D Shape and Motion of Fast Moving Objects", NeurIPS 2021, Code: https://github.com/rozumden/ShapeFromBlur
[2] Loper et al, "SMPL: A Skinned Multi-Person Linear Model"