Modular PySide6 Object Detection desktop App
Budget: $30 – $250 AUD
I need a Windows-only Python desktop application that is cleanly split into modules. The stack is fixed: PySide6 for the GUI, OpenCV for image handling, ONNX Runtime with the DirectML EP for inference, and DXCam for high-speed screen capture.
Core behaviour
• At start-up the app scans a “Models” folder, lets me pick an ONNX file, then instantiates the matching decoder class so any input / output tensor layout is handled transparently.
• Capture, inference, automatic target selection, PID calculations, the main UI, and plug-ins must each live in their own .py file; cross-talk happens only through well-defined interfaces.
• Target selection is automatic (based purely on the model’s predictions). Manual controls are not required now, yet keep the UI flexible enough that dropdown lists, selection boxes, and on-screen clicks could be wired in later without restructuring the code.
• A simulation mode must exist: when enabled, the computed control values move a dummy target rendered inside a test window that the app spawns itself. No process injection, memory scraping, anti-cheat work-arounds, mouse or controller automation, or any other form of unattended gameplay is permitted.
Deliverables (all must be met for acceptance)
1. Fully runnable project folder for Windows 10/11.
2. requirements.txt locked to known-good package versions.
3. step-by-step installation & launch instructions.
4. Well-commented source code, especially the model-adapter interface that links decoders to the core.
5. Example plug-in that does nothing except draw diagnostic overlays.
6. No forbidden automation or bypass methods present anywhere in the codebase.
The project is a priority; I’d like to see a first working build as soon as you can manage. Feel free to message me if any detail is unclear before you dive in.
Core behaviour
• At start-up the app scans a “Models” folder, lets me pick an ONNX file, then instantiates the matching decoder class so any input / output tensor layout is handled transparently.
• Capture, inference, automatic target selection, PID calculations, the main UI, and plug-ins must each live in their own .py file; cross-talk happens only through well-defined interfaces.
• Target selection is automatic (based purely on the model’s predictions). Manual controls are not required now, yet keep the UI flexible enough that dropdown lists, selection boxes, and on-screen clicks could be wired in later without restructuring the code.
• A simulation mode must exist: when enabled, the computed control values move a dummy target rendered inside a test window that the app spawns itself. No process injection, memory scraping, anti-cheat work-arounds, mouse or controller automation, or any other form of unattended gameplay is permitted.
Deliverables (all must be met for acceptance)
1. Fully runnable project folder for Windows 10/11.
2. requirements.txt locked to known-good package versions.
3. step-by-step installation & launch instructions.
4. Well-commented source code, especially the model-adapter interface that links decoders to the core.
5. Example plug-in that does nothing except draw diagnostic overlays.
6. No forbidden automation or bypass methods present anywhere in the codebase.
The project is a priority; I’d like to see a first working build as soon as you can manage. Feel free to message me if any detail is unclear before you dive in.