Computer Vision
Finding an Object
A detector is a picture of the thing you want and a window that walks the whole scene asking how much this looks like that. Drag the window anywhere to read its answer, or set it running and watch it sweep, filling in a score for every position and every size. The boxes come last, out of a threshold and an overlap rule.
the thing being looked forthe window right nowa box that survivedthe score, once that window has been asked
Stage 1 of 2: score every window · drag the window anywhere, Run sweeps it over everything
Found—
—
Look for
Search in
Score by
one size onlyhalf to one and a third
Values
windows asked—
score right here—
best so far—
multiplications—
What to observe
- This is stage 1 of a detector: score every window. Press Run and the same 16 × 16 comparison is asked at every position, filling in the map.
- One size and one handedness finds 4 of the 10 eyes. Ninesizes plus the mirror image find all 10, for 18 times the multiplications.
- Brightness divides by the spread of the window, so the faint robot's eyes still score high. The bare sum confuses faint with unlike.
- Lower the threshold: far more windows clear it than boxes are drawn, because each object answers at a pile of nearby windows. Thinning the pile is stage 2, Boxes and Overlap.
Shortcuts:space run/pause · s step · r reset · f fullscreen