Update model

[MateoLostanlen]

Our previous model was trained on images of size 1024. With the number of predictions that need to be made increasing on stations equipped with ptz cameras (smaller angle of view, therefore more images) and the reduction to 3 cpu, the pi doesn't have (almost) enough time to analyze everything. I'm also afraid of overuse. I therefore propose to switch to our best model in 640, reference 019aeef76fde49588d24fb167ba881a9.

We go from

On test test :

F1 : 0.916750 P : 0.934415 R : 0.899740 (threshold 0.1)

to

F1 : 0.890486 P : 0.933955 R : 0.850883 (threshold 0.1)

On test test :

F1 : 0.824006 P : 0.803797 R : 0.845258 (threshold 0.15)

to

F1 : 0.801754 P : 0.849442 R : 0.759136 (threshold 0.15)

I will re more training soon with 640 models only

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 92.38%. Comparing base (16fa010) to head (cac7c3a).

Additional details and impacted files

@@           Coverage Diff            @@
##           develop     #210   +/-   ##
========================================
  Coverage    92.38%   92.38%           
========================================
  Files            6        6           
  Lines          486      486           
========================================
  Hits           449      449           
  Misses          37       37           

Flag	Coverage Δ
unittests	92.38% <100.00%> (ø)

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.