[BUG]: Too many false positives

yunisguliyev commented 3 months ago

Before You Report a Bug, Please Confirm You Have Done The Following...

[X] I have updated to the latest version of the packages.
[X] I have searched for both existing issues and closed issues and found none that matched my issue.

DeepFace's version

0.0.92

Python version

3.12.4

Operating System

Macos

Dependencies

DeepFace

Reproducible example

from deepface import DeepFace

DeepFace.stream(db_path="database")

Relevant Log Output

No response

Expected Result

To detect my face

What happened instead?

Detected several faces that are not mine

Additional Info

So here I am using the Deepface for my auto-attendance project. I created dataset folder with the picture of 237 people(open source database + a picture of mine).

I tried the library in two different ways:

Was using the streamlit_webrtc library to get the image from the webcam via browser and after checking if there is face present in the frame(by using haar cascade algorithm) and in a positive case was passing the frame to the DeepFace.find() function. But unfortunately I was getting several face detection almost at every check, even tho I was the only person at the frame:

[ identity hash target_x target_y ... source_w source_h threshold distance 0 database/CFD-IM-688-322-N.jpg d14f0387f476a6bca7d28c7a0faef9cd77765470 668 1608 ... 336 480 0.68 0.440456 1 database/CFD-IM-696-001-N.jpg d4849a942f7b324da584f5ee27f6911fd881e55b 727 1619 ... 336 480 0.68 0.471839 2 database/CFD-MF-300-002-N.jpg 140d3d1b1e4d5902b9bf41523dd15069eaa01192 1820 1525 ... 336 480 0.68 0.555065 3 database/CFD-IM-715-013-N.jpg 2c391f02dd2627d0fb320065fe2c5c0c8e59c0d5 678 1599 ... 336 480 0.68 0.574597 4 database/CFD-IM-630-134-N.jpg 59aa030687fb7eb9d996f28946e37155b1c34f2b 753 1622 ... 336 480 0.68 0.580401 5 database/CFD-IF-633-012-N.jpg 2e276a78b10fa91d9e700f9d54ba9eae082b7d4c 1931 1462 ... 336 480 0.68 0.606692 6 database/CFD-IM-702-101-N.jpg ad68bb0719e09a2865ad09335473dad17cfd2dcb 1597 1636 ... 336 480 0.68 0.652795 7 database/CFD-IM-702-101-N.jpg ad68bb0719e09a2865ad09335473dad17cfd2dcb 655 1610 ... 336 480 0.68 0.654903

[8 rows x 12 columns]]

Then alternatively I tried using DeepFace.stream() function. I helped for a bit, but not much: 24-08-04 14:03:41 - Hello, database/yunis.png 24-08-04 14:03:41 - freezed 24-08-04 14:03:46 - freeze released 24-08-04 14:03:55 - Hello, database/CFD-IM-613-169-N.jpg 24-08-04 14:03:57 - freezed 24-08-04 14:04:02 - freeze released 24-08-04 14:04:05 - Hello, database/CFD-IM-607-018-N.jpg

Again, my face was the only at the frame.

serengil commented 3 months ago

Individual misclassifications are not bug. Deepface does not promise 100% accuracy. Its models precision and recall ratios on lfw are acceptable.

yunisguliyev commented 3 months ago

Any recommendations for troubleshooting ?

serengil / deepface