ozantezcan / bsuv-net-2.0 Goto Github PK

Hello !
First of all, thanks a lot for this great repository!
I am looking for a model to do background-subtraction in real time. This repository seems perfect, and I would like to improve it for my needs.
Some of the work I will do in my fork may be useful for other people, so I would like to know your politic regarding Pull requests and contributions. Are you interested by:

• having a dockerfile ?
• having new models ? For now only unet_vgg16 is implemented, but I would like to test other architectures that may run in real time on CPU (maybe something like unet_mobilenet). BTW, did you test other architectures ?
• having some proposal for code organization, cleaning, small changes, .... ?

Also for now the code doesn't have any license yet, so it is not Open Source (see https://docs.github.com/en/github/creating-cloning-and-archiving-repositories/licensing-a-repository#choosing-the-right-license). Are you planning to add a License soon ?

thanks a lot

Hey there,
it seems like the "currentFr" folder is missing in the linked GoogleDrive or can it be found elsewhere?

Hi
Are the input for f_score function correct?!
Shouldn't the predicted values ​​be rounded to calculate f_score ?

def f_score(true, pred):
    fn = torch.sum(true * (1 - pred))
    fp = torch.sum((1 - true) * pred)
    tp = torch.sum(true * pred)
    prec = tp / (tp + fp)
    recall = tp / (tp + fn)

is not possible when i run train.py?

I want you updated sources about it.

Thanks for the great work!

Is this the official implementation of BSUV-NET 2.0?
Are there pretrained BSUV-NET 2.0 and fast BSUV-NET 2.0 that I can quickly test on my own data?