Tips & troubleshooting
Get reliable recognition from the CNN embedding pipeline bundled with this demo.
Recognition pipeline
- Haar cascade finds a face region quickly.
- MobileNetV2 (ImageNet weights) embeds the crop.
- Cosine similarity compares against your stored mean embedding.
- Attendance rows store similarity, timestamp, and an optional snapshot path.
Common issues
Improve lighting, move closer, and match the pose used during enrollment. If similarity is slightly below threshold, ask an administrator to retake enrollment photos.
Browsers require HTTPS or localhost for camera APIs. Use Chrome/Edge/Firefox current versions and allow the prompt when opening Mark attendance.
This project stores snapshots under
data/snapshots for auditing. Treat the deployment like lab hardware—purge files when demos conclude.
Need admin support?
Enrollment changes, exports, and global attendance views live in the admin console.
Go to admin loginStudents: sign in with your roll number in the sidebar to unlock dashboards.
Session notice
Student sign-in uses roll numbers only—adequate for classroom demos, not production authentication.