Closed sepastian closed 4 years ago
The main problem seems to be the missing CUDA. Anyway GPU inside Container is not tested yet.
The fix is to use a CUDA-enabled docker base image, see https://github.com/OCR-D/core/pull/452 and https://github.com/OCR-D/core/pull/454
The fix is to use a CUDA-enabled docker base image, see OCR-D/core#452 and OCR-D/core#454
Does that actually work?
(I thought currently the only way to get GPU support was doing a CUDA installation in your OS and using ocrd_all natively.)
From what @sepastian writes in https://github.com/OCR-D/core/pull/452#issuecomment-594014728 I understand he got ocrd_anybaseocr running in Docker with CUDA (which must run on a CUDA-enabled host of course). I wanted to update https://github.com/OCR-D/core/pull/454 with your proposals in https://github.com/OCR-D/core/pull/454#pullrequestreview-406639090 so we can support -cuda
or -gpu
variants in ocrd_all eventually.
Inside the Docker container
ocrd/all:maximum
, the processorocrd-anybaseocr-dewarp
fails to run, because no GPUs can be detected.However, running
nvidia-smi
reveals that a GPU is available inside the Docker container.Does the system (inside the Docker container) require additional packages?