Shape from Blur: Recovering Textured 3D Shape and Motion of Fast Moving Objects

We address the novel task of jointly reconstructing the 3D shape, texture, andmotion of an object from a single motion-blurred image. While previous approachesaddress the deblurring problem only in the 2D image domain, our proposed rigorousmodeling of all object properties in the 3D domain enables the correct description ofarbitrary object motion. This leads to significantly better image decomposition andsharper deblurring results. We model the observed appearance of a motion-blurredobject as a combination of the background and a 3D object with constant translationand rotation. Our method minimizes a loss on reconstructing the input image viadifferentiable rendering with suitable regularizers. This enables estimating thetextured 3D mesh of the blurred object with high fidelity. Our method substantiallyoutperforms competing approaches on several benchmarks for fast moving objectsdeblurring. Qualitative results show that the reconstructed 3D mesh generateshigh-quality temporal super-resolution and novel views of the deblurred object.