Shape from projections via differentiable forward projector for computed tomography

In computed tomography, the reconstruction is typically obtained on a voxel grid. this work, however, we propose mesh-based method. For tomographic problems, 3D meshes have mostly been studied to simulate data acquisition, but not for reconstruction, which mesh means inverse process of estimating shapes from projections. paper, differentiable forward model that bridge gap between surfaces and optimization. We view projection as rendering process, make it by extending recent work in rendering. use proposed reconstruct directly Experimental results single-object problems show method outperforms traditional voxel-based methods noisy simulated data. also apply electron tomography images nanoparticles demonstrate applicability real