Loading Project Media
OFFLINE
Local-first inference
RUST
Native production runtime
HUMAN
Review remains in the loop
(Overview)
Hojirmovesacomputer-visionpipelinefromtrainingintoapractical,bilingualdesktopworkflow:thermalframe→orienteddetection→perspectiverectification→classification→qualitytriage→localreviewandexport.
(The Challenge)
Researchmodelsalonedonotmakeaninspectionproduct.ProductioninferencemustrunofflinewithoutPython,preserveorientedmodulegeometry,surfaceuncertainoutputsforhumanreview,andremainusableinbothEnglishandPersian.
(The Solution)
ModelstrainedinPyTorcharedeployedthroughONNXintoanativeRustruntime.ATauriandReactinterfacemanageslocalhistory,review,triage,andreportingwhilekeepingtheoperator—notthemodel—incontroloffinalinterpretation.
- 01
DevelopedtheorienteddetectionandclassificationpipelineforthermalUAVimagery.
- 02
DeployedtrainedmodelstoONNXandintegratedthemintoaPython-freeRustinferenceruntime.
- 03
Builtperspectiverectification,qualitytriage,andreview-readyresultprocessing.
- 04
Connectedinferencetoanofflinedesktopworkflowwithlocalhistory,export,andEnglish/Persianinterfaces.
Built with
PythonPyTorchRustTauriReactONNXONNX RuntimeYOLO OBBEfficientNetV2
Haveaprojectinmind?
amir@akbarpour.dev