search

Qengineering / Face-Recognition-Jetson-Nano

Recognize 2000+ faces on your Jetson Nano with database auto-fill and anti-spoofing
https://qengineering.eu/deep-learning-examples-on-raspberry-32-64-os.html
BSD 3-Clause "New" or "Revised" License
112 stars 32 forks source link

Low Accuracy #8

Closed AbhinandanVellanki closed 2 years ago

AbhinandanVellanki commented 3 years ago

Is there any way i can increase the accuracy of the face recognition? I noticed that while cropping out the detected face, prior to recognition, some of the face features are cropped out.

In most of the images, ears, or even a fraction of the face itself is not included in the cropped image. How can I increase the size of the cropped image so that all the face features are used to calculate the encoding and potentially improve the accuracy?

Qengineering commented 3 years ago

Dear @AbhinandanVellanki,

As you can see at line 330 in main.cpp, the boundaries of a face is given in the Faces[i].rect. This rect is filled by the face detection at line 284 or 286. With care you could enlarge the rect somewhat. Don't cross borders like x<0 or x>ImageWidth. Keep your landmarks at the location.

Qengineering commented 2 years ago

This issue has been automatically marked as stale because it has no recent activity. It will be closed if no further activity occurs. Thank you.