Reconsider max tile dim restriction

[alanocallaghan]

Consider unlimited tile sizes for those so bold as to request them

[petebankhead]

It isn't clear to me if you're referring to the UI or scripting.

@ThibautGoldsborough would know, but I have a vague memory that there is a limit to the number of predictions that can be made on a single tile - due to each pixel needing to be assigned an integer label and there being a risk of have too many labels for the int type. So we didn't want to permit unlimited tiles for that reason.

[alanocallaghan]

Either, there is a hard cap defined in both

...sure, but I would be inclined to let people run into that problem if they are lucky enough to have the vram to do it. Also for sparse tissue you could run a very large tile without hitting the integer limit?

[petebankhead]

Found the reference to the issue:

qupath-extension-instanseg/src/main/java/qupath/ext/instanseg/ui/InstanSegController.java

Lines 715 to 722 in 00ccede

	private void configureTileSizes() {
	// The use of 32-bit signed ints for coordinates of the intermediate sparse matrix *might* be
	// an issue for very large tile sizes - but I haven't seen any evidence of this.
	// We definitely can't have very small tiles, because they must be greater than 2 x the padding.
	tileSizeChoiceBox.getItems().addAll(256, 512, 1024, 2048);
	tileSizeChoiceBox.setValue(InstanSegPreferences.tileSizeProperty().getValue());
	tileSizeChoiceBox.valueProperty().addListener((v, o, n) -> InstanSegPreferences.tileSizeProperty().set(n));
	}

I don't think we should go bigger in the UI unless we've a clear need.

We could increase MAX_TILE_DIMS if we're able to determine that

1. it helps
2. it doesn't cause catastrophic problems when overflowing

I'm not entirely sure, but I think we need W x H x N < 2,147,483,647 where N is the number of seeds (?) and W and H refer to the dimensions of the binary mask generated for each seed.

Since √2³¹ is about 46,340, and we still need to divide that by N to get a sensible width or height, we really can't permit a very large number of seeds or instances within a tile - so problems could happen relatively quickly as the tile size increases.

[alanocallaghan]

"catastrophic" here refers presumably to tiles failing, which is perfectly fine imo for a scripting api where somebody is requesting a large tile size

[petebankhead]

Tiles failing would be ok - tiles giving incorrect results or blocking for ages because it's generating far too many objects would be much worse.

So we should check the consequences before making the change.