upgrade pytorch

[geoHeil]

recent versions of pytorch are more compatible with a wider variety of operating systems and cpu architectures. For example your specific version relies on an old version of torch-vision which only works on linux, not osx.

Do you have any plans to upgrade the pytorch version to something more recent?

[lucas-morin]

Hello @geoHeil,

We use a specific version of pytorch to make sure PaddleOCR works on GPU. PaddleOCR is used to:

1. Recognize abbreviations.
2. Pre-process images by removing possible captions.

If you want to use a different torch or torchvision version, you can try to edit it directly in the setup.py, here for GPU or here for CPU.

If you only need the CPU version, it should probably work. For GPU, you would need to refer to the PaddleOCR installation. It was only tested on x86_64, Linux Ubuntu 20.04, CUDA 11.7, and CUDNN 8.4.

To speed up the CPU version, you could also disable the caption removal with PaddleOCR here, by changing --remove-caption to --no-remove-caption. (The counterpart would be a minor loss of prediction quality.)

[geoHeil]

see #18