Performance limits of Annotorious + OpenSeadragon for large polygon annotations (10k+ points)

[danicabCVB]

Hi! I’m evaluating Annotorious (actual v3) on top of OpenSeadragon for a project that involves very large polygon annotations. Some polygons can have more than 10,000 points, and I’m running into a major performance issue:

• Creating the annotation is fine.
• But selecting the annotation becomes so slow.
• From what I can see, when an annotation is selected Annotorious updates the DOM with all polygon points (circles), which results in a huge number of DOM nodes being rendered at once. There the WebGL canvas does a good job, but the SVG polygons created stucks more than predicted.

Before I start exploring workarounds, I’d like to understand the current limits and expectations of Annotorious regarding very large annotations:

1. Is there an intended or practical upper limit for polygon size (number of points)?
2. Is the slow selection behavior expected when dealing with thousands of points?
3. Would you recommend a different approach?
4. Is there any planned support for more efficient rendering of large shapes (maybe in v4)?

Any guidance or best practices for handling large geometries would be greatly appreciated. Thanks!

[rsimon]

Hi! Yes, what you're seeing is the current behavior in v3.

• For viewing, Annotorious should generally handle mid-/large polygon datasets well (depending on your definition of mid/large :-). That includes many polygons as well as individual polygons with many corners. The reason, as you say: rendering is done in WebGL.
• On selection, Annotorious creates an editable SVG "copy" of the shape with draggable corner handles. At that point you inevitably hit the performance ceiling of the browser's SVG engine.

If you only need to select but not edit polygons, setting userSelectAction to SELECT would avoid the problem, because no SVG copy is created.

If editing is needed: at the moment there's no alternative path in Annotorious that avoids the SVG layer. Implementing a full WebGL/Canvas-based editing UI is certainly possible in principle. But it would be a bigger lift and isn't currently planned for v4. (One of the main changes is a switch from PixiJS to DeckGL for rendering, possibly introducing vector-tile-based rendering for very large datasets. But the editing tools would likely remain SVG for now.)

That said, if WebGL-native editing for very large geometries is something your project would benefit from, feel free to reach out. It’s the kind of feature that could be explored with dedicated funding or sponsorship. Hope this helps—and happy to answer follow-up questions!

[rsimon]

FWIW: the planning issue for Annotorious v4 includes a bit of info on how the next version will boost performance.