lodestar tutorials: added dataset download

[Pwhsky]

No description provided.

[mirjagranfors]

First one (231D) looks good. The dataset that is downloaded is a lot larger than the one you show (1764 frames instead of 363). I don't know if we should run all tutorials before uploading to avoid the outputs being different from what you get when you actually run the code?

In the second and third one, you have to check the downloading of the dataset. you download a different one than you want to use later.

In the third one, 231F, After the 2. Load data, you accidentally try to load tif_dir instead of just dir.

[Pwhsky]

Thank you for the critical feedback, I think it's best if we reduce the total dataset to a subset for the large case (i.e 363 frames) and keep the others the way they are, I have added a line in the BF-C2DL-Huh7 example that does this with

movie = movie[:363]

[mirjagranfors]

It looks quite good now!

If you only keep the first 363 frames, it won’t really change the training (since you only use one crop anyway), but it will affect the final plot. When only using the first 363 frames, the last plot will then only show four cells, which makes the result look less impressive, and you probably have to tune alpha and the cutoff.

Also, I think this part:
dataset_path = "cell_detection_dataset"
if not os.path.exists(dataset_path):
should include the dataset name, e.g.
dataset_path = "cell_detection_dataset/PhC-C2DL-PSC"
Otherwise, once the main folder is created, the other datasets won’t be downloaded when running the other tutorials.

I’d suggest running the script once before uploading. That way the outputs will match what users get when following the tutorial, and it also helps catch small issues like the ones I’ve mentioned.