Why Are They Implemented Differently?
BugSpot uses OpenCV because:
Python environment, OpenCV is a standard dependency
MOG2 is battle-tested, handles edge cases well
No performance constraints (desktop/server)
Trapjaw uses custom C/Metal because:
iOS deployment — cannot bundle OpenCV (massive binary, licensing)
Zero-dependency mandate — must build on Apple frameworks only
Real-time performance — Metal GPU shaders need a custom implementation
Memory control — fixed K=3 is deterministic memory usage per pixel
Options:
- Add Adaptive Component Selection to TrapjawChanges:- Implement fitness sorting (weight / variance) for BG classification- Add component creation/destruction logic (or at least "disable" unused components)
- Change BG classification from weight > 0.10 to cumulative weight > threshold
Pros: Closer to MOG2 semantics
Cons: Significant code changes to C + Metal shaders. Risks breaking GPU path.
- Port OpenCV's MOG2 Algorithm to C/Metal
Changes:
- Literally reimplement OpenCV's MOG2 in C and Metal
- Match every detail: fitness sorting, adaptive K, bootstrap, etc.
Pros: True 1:1 parity
Cons: Massive engineering effort. ~1000+ lines of C/Metal. Needs extensive validation.
Why Are They Implemented Differently?
BugSpot uses OpenCV because:
Python environment, OpenCV is a standard dependency
MOG2 is battle-tested, handles edge cases well
No performance constraints (desktop/server)
Trapjaw uses custom C/Metal because:
iOS deployment — cannot bundle OpenCV (massive binary, licensing)
Zero-dependency mandate — must build on Apple frameworks only
Real-time performance — Metal GPU shaders need a custom implementation
Memory control — fixed K=3 is deterministic memory usage per pixel
Options:
Pros: Closer to MOG2 semantics
Cons: Significant code changes to C + Metal shaders. Risks breaking GPU path.
Changes:
Pros: True 1:1 parity
Cons: Massive engineering effort. ~1000+ lines of C/Metal. Needs extensive validation.