Face registration works correctly in debug builds but fails in release builds during the embedding generation stage.
Expected Behavior
User captures face image
Face is detected
MobileFaceNet generates 192-dimensional embedding
Embedding is stored locally
Registration completes successfully
Actual Behavior
User captures face image
Face detection succeeds
Embedding generation fails
Registration process stops
User cannot complete enrollment
Impact
The authentication workflow depends on successful face registration. Since embeddings are not generated, verification cannot proceed.
Evidence
Observed error message:
Embedding Failed
What Works
Camera access
Face detection
Liveness detection workflow
Model loading in debug builds
Verification logic when valid embeddings are available
What Fails
Embedding generation in release APK
Registration completion in release APK
Suspected Causes
TensorFlow Lite release build optimization
Asset packaging differences
Native module behavior differences between debug and release builds
R8/ProGuard configuration
Release-specific inference pipeline issues
Status
Partially Resolved
Debug build functions correctly.
Further investigation is required to achieve consistent behavior in release builds.
Lessons Learned
Developing offline AI applications on mobile devices requires extensive testing across both debug and release environments. Functionality that works in development builds may behave differently after release optimizations are applied.
Face registration works correctly in debug builds but fails in release builds during the embedding generation stage.
Expected Behavior
User captures face image
Face is detected
MobileFaceNet generates 192-dimensional embedding
Embedding is stored locally
Registration completes successfully
Actual Behavior
User captures face image
Face detection succeeds
Embedding generation fails
Registration process stops
User cannot complete enrollment
Impact
The authentication workflow depends on successful face registration. Since embeddings are not generated, verification cannot proceed.
Evidence
Observed error message:
Embedding Failed
What Works
Camera access
Face detection
Liveness detection workflow
Model loading in debug builds
Verification logic when valid embeddings are available
What Fails
Embedding generation in release APK
Registration completion in release APK
Suspected Causes
TensorFlow Lite release build optimization
Asset packaging differences
Native module behavior differences between debug and release builds
R8/ProGuard configuration
Release-specific inference pipeline issues
Status
Partially Resolved
Debug build functions correctly.
Further investigation is required to achieve consistent behavior in release builds.
Lessons Learned
Developing offline AI applications on mobile devices requires extensive testing across both debug and release environments. Functionality that works in development builds may behave differently after release optimizations are applied.