YOLO — payslip field detector (docverify)

Finds the 10 fields of a French payslip so that an edit found by the tamper model can be pinned to net pay or gross salary instead of "pixels at (812, 640)". In docverify, an edit inside a money field is what turns a page into high risk.

Part of docverify: project · tamper model

Copy-move on net pay — field boxes and the edit

Classes

net_paye (net pay) · brut (gross salary) · salaire_base (base salary) · net_avant_impot (net before tax) · cotisations (total contributions) · iban · employee · employer · siret · period

Results

250 synthetic test pages, scans and phone photos mixed.

mAP50 mAP50-95
All fields 0.995 0.975
Weakest field (employee name) 0.995 0.943

Per-field mAP

19 ms per page on a T4; part of the 6.7 s full pipeline on 4 CPU threads.

Limits

Synthetic French payslips only (generated, fake people and companies); other layouts need their own field list.

License

AGPL-3.0, inherited from Ultralytics YOLO. The rest of docverify (code, tamper model) is Apache-2.0.

Downloads last month
5
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support