✳baebugs. / VISION LAB
SOURCE-BOUND COMPUTER VISION / V1

Every box has a source.

Explore recorded bounding-box predictions on real photographs of 3D-printed insect toys. Compare models, inspect confidence and coordinates, and follow each image back to its immutable evidence leaf.

RECORDED INFERENCE · NOT LIVE CAMERA · NOT BIOLOGICAL GROUND TRUTH

Detection workbench

Toggle the model layer, choose a photo, and select any box for its source identity.

See validation boundaries ↗
Model overlays
ORIGINAL SOURCE PHOTO + LIVE SVG BOX OVERLAY
3D printed insect toys used for visual detection comparison

Inspect, then contest.

Every published image and model proposal is bound to a source file hash. The viewer recomputes the derivative image and detection catalog digests when requested. Distinct hashes identify exact content; they do not establish accuracy or ownership of third-party models.

Source image SHA-256
—
Recorded prediction data
—
Published graph root
—
License / origin
—
Open original sourceView graph leaf

What this demonstrates

Two existing toy-photograph model comparisons, recorded YOLOE and OWLv2 boxes, a recorded ORDER classifier on some OWLv2 proposals, exact-byte image/source relationships and a human-inspectable interface. No detector weights are served; no new inference runs in the browser.

What remains unknown

No expert-confirmed taxonomy or independent accuracy benchmark for these images, no calibrated live-webcam pipeline at this website, no evidence that a robot recovered any specimen, and no new signed/Merkle Mountain Range commitment. Ultralytics weights and OWLv2 have separate upstream terms.

Remix the camera-intake protocol ↗