In theory the face recognition is quite accurate. What do real live examples show?
How to find the settings that work best for your pictures and how to apply them?
(Please connect if you want to see the video.)
00:20 Mode Probe explained in short
02:40 Pages - help - links to the pages thresholds and probe
03:20 Page - thresholds - distance metrics
03:45 Page - probe - step-by-step instructions
04:10 Start the tutorial with a fresh system (no pictures yet)
04:25 Upload pictures
04:40 Mount cloud files via webDAV (to later save CPU and memory on the server)
06:15 Admin - block Python scripts on the server
06:40 Admin - activate all detectors, models, distance metrics
07:35 Settings - activate all detectors, models, distance metrics
09:10 Run the Python script on your computer to detect faces
09:50 Watch the log file to see what is going on
10:50 Addon - name faces
11:35 Upload more pictures - every picture contains just one face of the persons we named before
13:45 Upload more pictures - no picture does contain a face of the persons we named before
15:15 Start probe - via web browser (1st way)
15:40 Start prove - via terminal on your computer (2nd way)
15:45 Results of probe - file
probe.csv - interpretations
21:50 Results of probe - file
statistics.csv - look at internal data of the face recognition
24:30 Effectiveness of face recognition - real live
25:15 Effectiveness of face recognition -
the theory25:15 Why the feature "probe" was implemented - compare the theory against real live performance
26:20 Results of probe - find the combination that works best for your real live pictures
29:15 Apply the best combination - thresholds
30:00 Apply the best combination - settings
30:40 (Admin - activate the detectors, settings and distance metrics)
31:10 Remove results for detectors and models we do not want to use anymore (blocked)
31:50 (Admin - allow the python scripts)
32:20 Remove results for detectors and models we do not want to use anymore (again)
32:40 Addon - start the face recognition again using your optimized settings
33:20 File statistics.csv - show that unused detectors and models where removed
33:43 The end