Pytorch-yolov3: why noobj_scale/obj_scale = 100 ??

Created on 16 Sep 2019  路  2Comments  路  Source: eriklindernoren/PyTorch-YOLOv3

why noobj_scale/obj_scale = 100 ??

picture CaoWGG

Most helpful comment

When you train 1 epoch or more, you can find:
loss-obj : 1.95
loss-noobj: 0.01
loss-noobj x 100: 0.87
So the scale factor balances the two loss.

picture ujsyehao on 13 Nov 2019
👍3

All 2 comments

When you train 1 epoch or more, you can find:
loss-obj : 1.95
loss-noobj: 0.01
loss-noobj x 100: 0.87
So the scale factor balances the two loss.

ujsyehao picture ujsyehao on 13 Nov 2019
👍3

@ujsyehao
Thank you for your answer

CaoWGG picture CaoWGG on 13 Nov 2019
Was this page helpful?
0 / 5 - 0 ratings

Related issues

a954217436 picture a954217436  路  4Comments
qilong-zhang picture qilong-zhang  路  5Comments
andohuman picture andohuman  路  3Comments
qianwangn picture qianwangn  路  3Comments
hezhangubc picture hezhangubc  路  4Comments