Python API usage question

[pr4deepr]

Hi
I have a question about BiaPy programmatic use in Python.
Looking at your page here, I can sees we can create and run BiaPy as a job by passing a yaml configuration file. Is there flexibility to pass a numpy array directly to BiaPy for inference?

Cheers
Pradeep

[danifranco]

Hi, we were thinking on implementing that but currently it's not possible. I will work on it today so it is available for the next release

[danifranco]

Hello! It took me more than expected but I have upgraded the BiaPy's Python API and now you can predict new images providing a numpy array as you requested. I added also more features to complete a bit more the options one might need when incorporating BiaPy into other code. Here is the documentation also updated: link

TLDR:

from biapy import BiaPy, build_config

# Load a model
biapy = BiaPy.load_workflow_from_model(...)
biapy.predict(image)

There are also other utilities you can use to print workflow information, such as:

biapy                              # short summary (workflow, dims, patch size, model, device...)
biapy.print_train_info()   # training overview: model, patch, epochs, LR, optimizer, augmentations, files
biapy.print_test_info()    # test overview: data path, GT, overlap/padding, post-processing, files

You can also predict all the images in a folder:

biapy = BiaPy.load_workflow_from_model(..., result_dir="/data/user/path")
biapy.print_test_info()
biapy.test()

I tested it but maybe there is an edge case I couldn't think on. I hope you find no errors and you like the new possibilities now! :D

PD: @lenkaback I mentioning you here cause you were also requesting a feature like this before. After thinking a bit I made it possible here. Hope you like it as well!