REVYRvehicle studio with a car and analysis interface.
REVYRvehicle scanREVYR· Vehicle evidence

Automotive vision/damage evidence

REVYR

A working browser proof that geometrically registers issue and return photos, locates new damage proposals and verifiably transfers each proposal to a human.

Working vehicle damage proof

The case in 30 seconds

REVYR

Problem
A working browser proof that geometrically registers issue and return photos, locates new damage proposals and verifiably transfers each proposal to a human.
Solution
From issue and return photo to registration, local damage detection, independent verification and human inspection.
Role and focus
Damage vision, registration
Status
Working vehicle damage proof
Result
Platform, demo, evidence suite

Live lab / try it yourself

Compare issue with return

Using two vehicle photos, check the geometric alignment and assess which new damage proposals deserve human investigation.

01 / InputBefore and after photoYou can change or choose this yourself
02 / SystemRegistration, detection and verificationImage registration · Computer vision · ONNX inference
03 / EvidenceA verifiable damage proposalThis makes the operation visible
InputIssue + return
RegistrationSIFT + RANSAC
DetectionLocal ONNX model
DecisionHuman control

Frozen test suite

Four ports, one fixed case

The proof measures reproducibility within one controlled image pair. The results are convincing for that suite, but deliberately not a fleet-wide accuracy claim.

01 · Damage variants10/10

Dent and scratch localized, without additional notification.

02 · Zero checks10/10

Photometric variants deal exactly zero new damage.

03 · Registration10/10

Perspective, rotation and translation remain within the pre-selected limit.

04 · AI Verifier10/10

Two label-free proposals independently confirmed ten times.

Geometry before conclusion

Place the same panels on top of each other

Differences in location, perspective and cropping may otherwise appear as apparent damage. Drag to compare issue and registered return.

Registered issue photo of the vehicleRegistered return photo of the same vehicle
IssueReturn

Chain of custody

No magical look

REVYRseparates recording quality, geometric registration, local model proposal, independent verification and human decision.

01 / CAPTURE

Test photo first

Angle, sharpness, opacity and light determine whether comparing is useful.

02 / REGISTER

Align panels

Vehicle anchors bring the same parts into one comparison space.

03 / PROPOSE

Local AI highlights

The ONNX model represents class, score and region without evaluator or target coordinates.

04 / VERIFY

Man decides

A confirmed region remains a concern, not an automatic claim or liability.

REVYRevidence board with before, after and two suggested damage zones
Proof with reach

40/40 is strong and limited

The current suite proves that this fixed case reproducibly passes through damage, zero, registration and verification gates. The next credible step is a blinded fleet pilot with other vehicles, angles, weather conditions and damage sizes.

  • 30 unique local entries with hash and pixel checking
  • Raw inference saved before scoring
  • Evaluator not included in the model
  • No independent claim or liability decision

Open the platform

From photo to inspection

View the complete product site, run the fixed demo or access the reproducible evidence file with the test ports and limits.

More concept work

Other working proofs

From care vision and security evidence to voice operations.