Parallelized, GPU-accelerated, and differentiable digitally reconstructed radiographs in Julia.
NOTE: THIS PACKAGE IS NO LONGER MAINTAINED.
For similar functionality, use DiffDRR
instead (https://github.com/v715/DiffDRR).
DDRs.jl
was an experimental digitally reconstructed radiograph (DRR) generator in Julia. It's aim was to be a GPU-accelerated, auto-differentiable renderer (a la RayTracer.jl) for medical images, but challenges using CUDA.jl
and Zygote.jl
made it difficult to reach a minimal viable prototype.
๐ซ DRRs.jl
inspired a PyTorch version that is fully functional and publically available at DiffDRR
๐คท๐พโโ๏ธ Lessons learned from DiffDRR
may be ported back to Julia at some point, but having to write custom differentiation rules is really annoying (compared to PyTorch's autograd, which just works ๐)