Is there an existing issue for this?
What happened?
ObjectClassifier and FaceDetector are initialized outside the try...finally block in
backend/app/utils/images.py. This can lead to resource leaks when initialization or cleanup fails.
Details
Current pattern:
object_classifier = ObjectClassifier()
face_detector = FaceDetector()
try:
...
finally:
object_classifier.close()
face_detector.close()
Problems:
If FaceDetector() raises during init, finally is never executed → ObjectClassifier is not closed.
If object_classifier.close() raises, face_detector.close() is skipped.
Impact:
- ONNX session / memory leaks
- Potential file handle leaks
- Increased risk of backend instability during AI Tagging
Proposed Fix
Refactor to:
- Initialize both objects as None
- Move initialization inside try (or use nested try...finally)
- Guard each close() call (e.g. if obj is not None)
- Optionally isolate close() calls to avoid cascading failures
Testing
Manual: verify AI Tagging still works; simulate model load failure
Automated (optional): mock init failure and assert close() is called
Request:
Happy to submit a PR with the fix + test if this approach is acceptable.
Record
Is there an existing issue for this?
What happened?
ObjectClassifier and FaceDetector are initialized outside the try...finally block in
backend/app/utils/images.py. This can lead to resource leaks when initialization or cleanup fails.
Details
Current pattern:
Problems:
If FaceDetector() raises during init, finally is never executed → ObjectClassifier is not closed.
If object_classifier.close() raises, face_detector.close() is skipped.
Impact:
Proposed Fix
Refactor to:
Testing
Manual: verify AI Tagging still works; simulate model load failure
Automated (optional): mock init failure and assert close() is called
Request:
Happy to submit a PR with the fix + test if this approach is acceptable.
Record