Darknet Letters

Blender Game Engine …. Intelligence Artificielle

font_0_h 0.37 0.31 0.18 0.18
font_0_l 0.07 0.27 0.13 0.13
font_0_t 0.44 0.14 0.1 0.1
font_0_C 0.54 0.49 0.13 0.13
font_0_O 0.27 0.95 0.21 0.21

De l'Intelligence artificielle pour reconnaitre des lettres dans différentes polices, et en faire des cauchemars et de la musique en midi.

Les sources sur GitHub

darknet-letters

Ressources

https://github.com/AlexeyAB/darknet

How to detect your custom objects

Pre-trained weights

wget -c https://pjreddie.com/media/files/darknet53.conv.74

Fichier *.cfg

For training for both small and large objects use modified models:
Full-model: 5 yolo layers:

https://raw.githubusercontent.com/AlexeyAB/darknet/master/cfg/yolov3_5l.cfg

change line batch to batch=64
change line subdivisions to subdivisions=8
change line max_batches to (classes*2000), soit 800 000
change line steps to 80% and 90% of max_batches, f.e. steps=640000,720000
 
change line classes=80 to your number of objects in each of 3 [yolo]-layers:
        Line 610
        Line 696
        Line 783
        classes=400
 
change [filters=255] to filters=(classes + 5)x3 in the 3 [convolutional] before each [yolo] layer
        Line 603
        Line 689
        Line 776
        filters = (classes + 5)x3 = (400 + 5) * 3 = 1215
 
Distinction main gauche main droite
ligne 17    flip = 0

ia, sb