Grade GUI annotations from medical imaging software
Each screenshot shows one red box and three instructions an AI wrote for it. Tell us whether the box fits and whether each instruction points at exactly that box. A session is 20 screenshots (about 5 minutes); do as many sessions as you like.
How to grade
The box
- Correct — tightly encloses one clickable element (button, menu item, field, tab, icon).
- Loose — the right element, but the box is clearly too big, too small or shifted.
- Wrong — not a sensible single target (empty space, a whole panel, several elements).
Each instruction (functional, visual, spatial)
Judge it against the box as drawn, even if the box is wrong.
- Yes — someone following it would click exactly the boxed element.
- Ambiguous — it could just as well mean another element on the screen.
- No — it points somewhere else, or says something false about the screen.
Example: box tightly around the "Try Advanced Segmentation Plugins" button, instruction "Click the top button in the Models List panel" → box correct, instruction yes. The same box drawn around the whole Models List panel → box wrong; "Try the advanced segmentation plugins" → no (the panel is not that button).
Keys: box 123 · functional QWE · visual ASD · spatial ZXC (yes / ambiguous / no) · O box on/off · Enter next.
We store only the name you type and your answers, to analyse annotation quality.