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

Question about how to install inplace-abn

FFFFlamingo opened this issue · comments

Thanks for your great work!
How do I install the corresponding version of Inplace-ABN on Windows?
'Pip install inplace-abn' doesn't work.Neither does 'git clone' the package and then 'python setup.py install'.
I look forward to your reply.

Hello. Please refer to this repository for the install instruction: https://github.com/mapillary/inplace_abn

I tried to follow their instructions but they throw errors.
https://pastebin.com/BqUnhDSW

It seems you have trouble compiling their library. Can you please check if the CUDA and GCC you are using match the requirement?
Did you try to install the library from pip?

Yes all requirements are satisfied. "NOTE 2: we are only able to provide support for Linux platforms and CUDA versions >= 10.0" I am using CUDA 11.7.
I tried using both pip and from source as given in their instructions. Both give the same errors.
Can this repository code be updated to account for new changes: using torch.amp and from torch.nn.parallel import DistributedDataParallel as DDP

Take a look at Mibv2 here: https://github.com/fcdl94/MiBv2/

Thanks for the new implementation. I tried to create an environment using requirements.txt but the inplace-abn package doesn't correctly get installed and I believe the model uses it. Is there a way to do without inplace-abn or can you push the inplace-abn version which works to the repo?

Inplace-abn installation is not under my control 😅 Please if you are not able to install it, check their repository. In-place ABN is mandatory for the project, there is no specific version for it (I think only one was released).

Hi @fcdl94

what is the CUDA version you used for your code?

It was CUDA 10.2.