yolact icon indicating copy to clipboard operation
yolact copied to clipboard

Published 20 hours ago • verified publisher icon dbolya

Train the final layer

Open monjurulkarim opened this issue 5 years ago • 2 comments

Hi, Thanks for this amazing repository. I am trying to train only the final layer ( network head) with my own very small dataset. How can I do that? Can you please help?

monjurulkarim avatar Feb 13 '20 00:02 monjurulkarim

You can do this by following the fine tuning procedure (#36) and then here: https://github.com/dbolya/yolact/blob/f54b0a5b17a7c547e92c4d7026be6542f43862e7/yolact.py#L628 replace that with

p = pred_layer(pred_x.detach())

dbolya avatar Feb 19 '20 00:02 dbolya

Hi, I am trying to train on a very small dataset consisting of only 10 train images and 5 val images. When calculating mAP it is remaining everything 0. Is it because the dataset is relatively small? Thanks

edrica-caruana avatar Apr 06 '22 16:04 edrica-caruana