[ziw-liu]

Adds a short description of the virtual staining workflow in its current state.

[mattersoflight]

LGTM

[mattersoflight]

This addresses #43.

[edyoshikun]

LGTM! Thanks @ziw-liu

[edyoshikun]

Maybe this is for a separate issue, but I had also trouble running the metrics/test command. I think this was just a lack of time on my side during the course to fully explore pytorch lightning framework. I think that if we run it together @ziw-liu this comment can be solved and we can add further documentation as separate PR.

[Soorya19Pradeep]

Thank for putting together this example @ziw-liu! It was very easy to navigate and documentation was great!
I was able to navigate through the solution and train a 2D HEK nucleus-membrane prediction model using uninfected cells in the TICM0036 dataset acquired on Dragonfly microscope. I was able to test the model by performing prediction on OC43-infected HEK cells at 48hpi.

I did make changes on the line:
https://github.com/mehta-lab/VisCy/blob/c13176d8d9b21cfc3a9b2ead913f748bc0dd775d/examples/demo_dlmbl/solution.py#L396

I changed this based on @ziw-liu 's suggestion.

phase2fluor_config = {
    "num_filters": [24, 48, 96, 192, 384],
    "in_channels": 1,
    "out_channels": 2,
    "residual": True,
    "dropout": 0.1,  # dropout randomly turns off weights to avoid overfitting of the model to data.
    "task": "reg",  # reg = regression task.
}

phase2fluor_model = VSUNet(
    architecture="2D",
    model_config=phase2fluor_config.copy(),
    loss_function=torch.nn.functional.l1_loss,
    schedule="WarmupCosine",
    log_num_samples=5,  # Number of samples from each batch to log to tensorboard.
    example_input_yx_shape=YX_PATCH_SIZE,
)