Giters

fcdl94 / MiB

Official code for Modeling the Background for Incremental Learning in Semantic Segmentation https://arxiv.org/abs/2002.00718

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

about pretrain model

ygjwd12345 opened this issue · comments

commented

Hi，
I want to reproduce your work.
but I get an error as follow
FileNotFoundError: [Errno 2] No such file or directory: 'pretrained/resnet101_iabn_sync.pth.tar'
can you tell me how to get it?

commented

I find it
thanks

commented

@ygjwd12345 Could you tell me how you resolved this issue?
I'm facing similar issue

commented

Hi @deepakksingh.
I've seen only now your comment.
That's the link: https://github.com/mapillary/inplace_abn#training-on-imagenet-1k (click on md5 to get the download page).
If you open the link, there's a table. The last row is the pretrained model you need, but in case you want to try different architecture, you can just download the pretrained and specify the correct backbone.

commented

@fcdl94 Thank you for your reply.
I did the same. I downloaded it from the Google drive and renamed it accordingly.

commented

@fcdl94 Thank you for your reply.
I did the same. I downloaded it from the Google drive and renamed it accordingly.

Hi~ I download corresponding pth.tar file for this project from inpace_abn repository. But the parameters' name doesn't match. Do you encounter this problem?

thanks~

commented

Hi!

Which parameters does not match? That's strange since I took their resent and deeplab implementation (have you specified the norm_act parameter maybe?)

commented

I faced the issue where "module" word was not matching. That happens when the model was trained and saved as per multi-gpu setting. So you may have to rename some parameter names to remove the "module" word

commented

Follow up : The renaming issue is mentioned here : #14 (comment)

commented

I faced the issue where "module" word was not matching. That happens when the model was trained and saved as per multi-gpu setting. So you may have to rename some parameter names to remove the "module" word

yes, it is the reason why not matched. ~ I didn't read README carefully. Thank you~

commented

Hi!

Which parameters does not match? That's strange since I took their resent and deeplab implementation (have you specified the norm_act parameter maybe?)

Thanks for your reply. It is the reason of "module." in pretrained model's state_dict's keys~