tests: unit tests do not check the outputs of models

[Mayukhdeb]

While the unit tests do cover the output types, they do not check if the outputs themselves are correct/incorrect.

To do:

• Use some loss function w.r.t ideal outputs and make sure the losses are low enough
• Make sure the output dtype matches the original (int, float32, float64)

[vrutikrabadia]

Hello @Mayukhdeb, for the lineage population model we can calculate loss from the ground truth population labels from the CSV. But, I don't understand how can we test the output of the segmentation model. One way I think could be running the model on the same image multiple times and calculating the mutual loss between them. If that would be helpful, I wish to work on the issue.

[Mayukhdeb]

Hi @vrutikrabadia.

For the segmentation model, I plan to execute the testing procedure by saving a few .npy files which would then be compared with the predictions with np.allclose().