Randomized Image Occlusion

Anki image occlusion - but you can't cheat.

An Anki add-on for studying labelled diagrams: anatomy cross-sections, the brachial plexus, histology slides, bones, ECGs, etc.

Built for Ezra - a medical student at the University of Glasgow, whose idea and request this whole thing grew out of. 💛 See Thanks.

The problem it fixes

Normal image occlusion hides a label on a diagram and asks you to recall it. It works - until it doesn't. Because the hidden box always sits in the same spot, your brain quietly starts taking a shortcut: you remember "the box in the top-left is the aorta" from where it is, not because you actually recognise the aorta.

That's spatial memory doing your revision for you. It feels like you know the diagram. Then the exam shows the same structure from a slightly different angle, and it's gone.

What it does instead

Every single time a card comes up, this add-on drops the answer box in a random place on the image and draws an arrow from the box to the real structure. There's no fixed position to lean on: you have to follow the arrow and genuinely identify what it's pointing at.

Same idea as image occlusion. Same one-click workflow. It just refuses to let you memorise the layout instead of the material.

The shuffling is purely visual: it never touches Anki's scheduling or your review history. Your stats and FSRS stay exactly as they were.

See it in action

Setting up the card maker

Entering the setup: navigating to the drop-down menu.

Labelling a card

Labelling a card: setting up markers for occlusion.

Moving card occlusion markers

Moving card markers: moving occlusion markers around the card.

Deleting card occlusion markers

Deleting card markers: deleting and reinstating individual occlusion markers.

Filling card occlusion metadata

Filling card metadata: filling in card occlusion metadata.

Demo test run

Reviewing a card: the prompt box lands somewhere new each time, with an arrow to the structure.

Extra study modes (optional)

You can leave everything on its sensible defaults, or turn on:

• Type the answer - type the structure's name and let Anki grade it, instead of flipping to reveal. (Note: if a label contains ::, {{ or }} - e.g. a C++/Rust name like std::vector - Anki's type grader compares against an escaped form, so use reveal or single-card mode for those labels.)
• Reverse cards - instead of "what is this?", get "where is the X?" and find it. Or both: still one card per structure, but which way it asks is re-rolled at random each review.
• Context labels - show the surrounding labels while you answer, the way "hide one, guess one" occlusion does.
• Single-card mode - put a whole diagram on one card that cycles through every label in a fresh random order each review, with a running counter.

Works in both light and dark mode, and on your phone (AnkiDroid / AnkiMobile) - the card carries everything it needs, so it renders even where the add-on isn't installed.

Requirements

Anki 23.10 or newer (desktop).

How to install it

The easy way - from AnkiWeb (listed as Randomized Image Occlusion):

1. In Anki, go to Tools → Add-ons → Get Add-ons….
2. Paste in the code 1836497069 and click OK.
3. Restart Anki.

You can also find the listing here: https://ankiweb.net/shared/info/1836497069.

Or install from a file (e.g. before it finished syncing to AnkiWeb):

1. Download randomized_occlusion.ankiaddon from the Releases page.
2. Open Anki and go to Tools → Add-ons → Install from file…, then pick the file you just downloaded. (On most computers you can also just double-click the file.)
3. Restart Anki.

Either way, you'll now see Tools → Randomized Image Occlusion… in the menu.

How to use it

1. Go to Tools → Randomized Image Occlusion….
2. Click Load image… and choose your diagram - or press Ctrl+V to paste one straight from the clipboard (see Pasting an image).
3. Click each structure on the image to drop a numbered marker, and type its label. Repeat for every part you want to learn. (Drag a marker to reposition it; click the × in the list to remove one.) Scroll to zoom in if you need to be precise - see Zooming in.
4. (Optional) Add a header, some back-of-card notes, pick a deck, and choose any of the study modes above.
5. Click Save. Your cards are added to the deck.

Then just review like any other Anki cards. Each time, the prompt box lands somewhere new with an arrow to the structure - so you're always answering "what is this?", never "what usually goes in this corner?".

Pasting an image

You don't have to save a picture to disk first. Press Ctrl+V, or click Paste image, and whatever is on the clipboard is used:

• a screenshot (Windows Win+Shift+S, macOS Cmd+Shift+4, most Linux screenshot tools) - saved as a PNG;
• an image copied from a web page or another app - kept in the format that app published, so a copied animated GIF stays animated;
• an image file copied in your file manager - used as-is, keeping its own name and bytes.

This works both when creating a card and when editing one, and Load image… still works exactly as before if you'd rather pick a file. Ctrl+V still pastes text wherever you are typing — the Header and Back extra boxes, and the label fields beside the image — so it only reaches for the clipboard's picture when you are not in a text field.

Replacing the image clears the markers you have placed, so if there are any you will be asked to confirm first.

Zooming in

Dense diagrams need a closer look, so the canvas zooms:

Left-clicking always places a marker, at every zoom level, so panning never drops one by mistake. Click a marker to select it before nudging. Zooming only changes what you see - marker positions are stored relative to the image, so they come out identical however far in you were when you placed them.

The zoom level is remembered for next time, and works the same when you come back to edit a card later.

From Anki's Add window

You can also make cards straight from Add: choose the Randomized Image Occlusion note type and click the Occlusion button in the editor toolbar to mark the image up on the canvas, then press Anki's Add. The internal fields (Image, Structures, Ordinals, TypeAnswer) are collapsed out of the way so you never have to touch them by hand.

Editing a card later

Made a typo, or want to nudge a marker? Open the Browse window, right-click the card, and choose Edit with Randomized Image Occlusion. The image and all its markers reappear on the canvas exactly as you left them - move them, rename them, add or remove structures, change the study mode, then Save. Zooming, nudging and pasting a replacement image all work here too, so a marker that ended up slightly off can be corrected precisely. Anki updates the card (and adds or removes cards if you changed the number of markers) in a single undo step.

Settings

Prefer different colours, a longer minimum arrow, or a different default mode? Everything is adjustable under Tools → Add-ons → Randomized Image Occlusion → Config. Each option is documented in config.md.

Thanks

This add-on exists because of Ezra!

For developers

The code lives in src/randomized_occlusion/. The core logic has no dependency on Anki and is fully unit-tested; only the thin editor and menu shells need Anki to run.

pip install -e ".[dev]"
pytest                       # Python tests (domain, config, note pipeline, fuzz)
node --test tests/js/*.test.js   # headless tests for the reviewer and editor JS
python mutate.py             # mutation testing (see below)
ruff check .                 # lint
mypy                         # type-check
python build.py              # writes dist/randomized_occlusion.ankiaddon

Coverage shows that a line executed. It does not show that any assertion depends on the result: a test can run a line, assert something true either way, and pass once the logic behind it is broken. mutate.py replaces one expression at a time and reports whether the suite fails. Nothing is written to the working tree; each mutant exists only inside the process that runs it.

python mutate.py               # the whole catalogue (several minutes)
python mutate.py --list        # what is in it, without running anything
python mutate.py marker.js     # filter by label or by file

It fails in four cases: a mutation nothing caught (the behaviour is not tested), an anchor that no longer matches its source (the code changed and the catalogue did not), a mutation listed as knowingly untested that a test did catch (the exemption should be removed), and a mutation that never reached the file at all (the result would say nothing either way).

Two checks run first, because a kill is only evidence if the suite failed for the reason claimed. Every mutant is compiled: one with a syntax error fails its whole suite at load, which would otherwise be recorded as a kill. Then a no-op mutation is applied to each target and must survive, which catches a suite that is already red, or a runner that cannot reach the file. tests/test_mutate.py revalidates the catalogue against the source on every ordinary test run.

The Python suite includes randomized property/fuzz tests (test_fuzz.py) that hammer the note round-trip and payload invariants, and the JS suite runs the reviewer's placement/RNG logic headlessly (Node's built-in runner, no deps).

License

Apache 2.0. See LICENSE.