cfg_yolo

Script to create your own .cfg to train a Darknet YOLO model

How to use

First clone the repo or download it by clicking here: click

In order to create the .cfg file for your custom YOLOv4 model run:

$ python cfg_create.py -i yolov4.cfg -c <number of classes to train> 

this will create a file named yolov4_custom.cfg in the same path as yolov4.cfg.

Example 2:

If you want to train a tiny-yolov3 to detect 6 different classes you can run:

$ python cfg_create.py -i yolov3-tiny.cfg -c 6

Arguments

argument name	default	Description
"-input" or "-i"		"path to the .cfg"
"-classes" or "-c"		"How many classes to detect"
"-num_images" or "-n"	6000	"OPTIONAL: Number of training images "
"-width" or "-wi"	416	"OPTIONAL: network size- width "
"-height" or "-he"	416	"OPTIONAL: network size- height "
"-batches" or "-b"	64	"OPTIONAL: batch size during train "
"-subdivisions" or "-sub"	32	"OPTIONAL: subdivisions during train "
"-no_flip"	True	"OPTIONAL: use -no_flip to set flip=0 "

only use -no_flip if you train the model to distinguish Left and Right objects as separate classes (left/right hand, left/right-turn on road signs, ...).

If you are in doubt about any parameter value you should check AlexeyAB darknet repo