Darknet: What are the parameters to change in yolov3.cfg file for training with our own dataset?

Created on 6 Feb 2019  路  1Comment  路  Source: pjreddie/darknet

Hi sir,
I have some doubt about Keras Yolov3.What are the parameters that I have to change for training Keras Yolov3 with my own dataset....
What are the changes that I have to make before converting yolov3.cfg & yolov3.weights to yolo.h5 file?
Recently I had made few changes in yolov3.cfg file.. ie, first I changed the number of classes to 1(I have only one class). Secondly I changed the anchors(I developed anchors from my own dataset using kmeans.py file found in qqwweee/keras-yolo3 github.I got 9 pairs of anchors(5,6, 6,8, 8,10, 10,12, 12,16, 17,22, 26,33, 45,57, 111,147)).
1.What are the other parameters that I have to change?

  1. In yolov3.cfg file I found a parameter called num (num=9).Is any change is required for this num parameter for training the model with my own dataset?
    3.Whether I need to make any changes in filters?If so can you please give me the actual mathematical equation for finding the number of filters?
    Please.... Help...

Most helpful comment

>All comments

Was this page helpful?
0 / 5 - 0 ratings

Related issues

AndyZX picture AndyZX  路  3Comments

TheHidden1 picture TheHidden1  路  3Comments

cadip92 picture cadip92  路  3Comments

ryuzakinho picture ryuzakinho  路  4Comments

groot-1313 picture groot-1313  路  4Comments