PytorchSSD icon indicating copy to clipboard operation
PytorchSSD copied to clipboard

Loss does not decrease for coco

Open strongwolf opened this issue 6 years ago • 3 comments

Have you ever trained on coco for these models? I try to train coco with original SSD. However, the loss keeps increasing especially for the localization loss from 3 to 30! The accuracy also decreases. I use the default config and everything behaves well on Pascal voc.

strongwolf avatar Aug 30 '18 08:08 strongwolf

I will try it later. Please keep pending.

lzx1413 avatar Aug 30 '18 08:08 lzx1413

Have you ever trained on coco for these models? I try to train coco with original SSD. However, the loss keeps increasing especially for the localization loss from 3 to 30! The accuracy also decreases. I use the default config and everything behaves well on Pascal voc.

Hello, I also meet this problem, have you solved it?

Chida15 avatar Aug 01 '19 07:08 Chida15

Have you ever trained on coco for these models? I try to train coco with original SSD. However, the loss keeps increasing especially for the localization loss from 3 to 30! The accuracy also decreases. I use the default config and everything behaves well on Pascal voc.

Hello, I also meet this problem, have you solved it?

The reason is that the augmentation operations have some errors that are not appropriate for some background images in coco.

strongwolf avatar Jun 16 '20 06:06 strongwolf