Credits

Whose work PanelProof stands on.

The public pictures the app learned from and was tested on, and the open software and typefaces it is built with.

Pictures

Read this first

A credit is not an endorsement

Nothing on this page means that any author, company or organisation named here supports PanelProof or is connected with it. None of them is a PanelProof customer.

Training and testing

InfraredSolarModules

InfraredSolarModules. Copyright (c) 2020 Raptor Maps, Inc. Released under the MIT License. Source: github.com/RaptorMaps/InfraredSolarModules. Paper: Millendorf, M., Obropta, E. and Vadhavkar, N., "Infrared Solar Module Dataset for Anomaly Detection", ICLR 2020 workshop on AI for Earth Sciences.

Raptor Maps published this dataset. Raptor Maps has no connection with PanelProof.

Used for training, tuning and the held back test. Our changes: its 11 fault classes grouped into 5, the pictures split into three parts, and copies with small changes in brightness, position and sharpness added during training. Where its pictures are shown they are enlarged.

Second test

sUAS Infrared Image Dataset for Photovoltaic Module Fault Classification

Schaefer, C. (2026). sUAS Infrared Image Dataset for Photovoltaic Module Fault Classification [data set]. University of Louisiana at Lafayette. Zenodo. doi.org/10.5281/zenodo.20860870. Licensed under Creative Commons Attribution 4.0 International.

Used for testing only. Our changes: each panel turned upright, its black corners cut away and the few corner pixels left filled from the nearest panel pixel, then resized to 24 by 40 pixels.

Real photos

UAV radiometric thermal imagery of PV module faults, sample dataset

Pratticò, D. (2026). UAV radiometric thermal imagery of PV module faults, sample dataset (1.8 MW plant, Catanzaro, Italy), version 1 [data set]. University of Reggio Calabria. Mendeley Data. doi.org/10.17632/5jkd8b3zh5.1. Licensed under Creative Commons Attribution 4.0 International.

Used for testing full photos. Our changes: the camera's readings read out of each file, white hot pictures drawn from them, panels cut out with grids, turned upright and resized, and the app's marks drawn on top. The original title uses a dash where this credit uses a comma.

Farm photos

Thermal PV Panel Detection and Fault Detection Dataset for UAV-Based Inspection

Christakakis, P., Pechlivani, E.-M. and Dimou, P. (2025). Thermal PV Panel Detection and Fault Detection Dataset for UAV-Based Inspection [data set]. Centre for Research and Technology Hellas. Zenodo. doi.org/10.5281/zenodo.16420123. Licensed under Creative Commons Attribution 4.0 International.

Used for testing how many panels the app flags. Our changes: single panels cut from the authors' boxes with a tenth trimmed off every side, turned upright and resized to 24 by 40 pixels.

Type and software

Typefaces

Three open typefaces

  • Apfel Grotezk by Collletttivo. SIL Open Font License 1.1.
  • Inter by The Inter Project Authors. SIL Open Font License 1.1.
  • Instrument Serif by The Instrument Serif Project Authors. SIL Open Font License 1.1.

Software

Open source libraries

LibraryByLicence
Next.jsVercel, Inc.MIT
ReactMeta Platforms, Inc.MIT
TypeScriptMicrosoft CorporationApache 2.0
three.jsthree.js authorsMIT
FastAPISebastián RamírezMIT
PydanticSamuel Colvin and contributorsMIT
UvicornEncode OSS Ltd.BSD 3-Clause
asyncpgMagicStack Inc. and contributorsApache 2.0
ONNX RuntimeMicrosoft CorporationMIT
PyTorchThe PyTorch authorsBSD style
NumPyThe NumPy developersBSD 3-Clause
SciPyThe SciPy developersBSD 3-Clause
scikit-learnThe scikit-learn developersBSD 3-Clause
PillowJeffrey A. Clark and contributorsMIT-CMU
ReportLabReportLab Inc.BSD
PostgreSQLThe PostgreSQL Global Development GroupPostgreSQL Licence

DJI and Zenmuse are trademarks of SZ DJI Technology Co., Ltd. The app reads DJI camera files with its own code and uses no DJI software. DJI has no connection with PanelProof.