similarity Implement multi-siam for segmentation

https://arxiv.org/pdf/2108.12178.pdf

Oct 19 '21 16:10 ebursztein

Hi , My name is Wansh Singh and i am an absolute beginner to this open source contribution stuff and want to begin my journey by working on this issue. So i just want to get assigned for this issue and i also need a quick guide on what i have to learn and use to solve these issue. So, @ebursztein can you provide me some more resources regarding this?

Jan 29 '22 12:01 Wansh619

@ebursztein please i want to know whether i will be assigned for this issue or not as i am really desperate to contribute

Jan 31 '22 03:01 Wansh619

Hi @Wansh619, let me look into it and see how we might either add it to the existing Contrastive model or if we might need a separate model. At first glance, it looks like they based their base model on BYOL, so we likely can use the SimSiam algorithm in the tf.sim contrastive model as it also uses the pairwise cosine distance and will remove the need for BYOL's momentum component.

However, I'll have to look into it more to see how we "extend the cross-view consistency framework to multi-instance circumstances". What are your thoughts?

Feb 01 '22 05:02 owenvallis

Actually i am very new i this open source stuff and i want to have a hewd start so i thought i should start with a easy issue and ask for help in the community as i am very very new to this .so can you guide me taht what i have to learn for this contribution?

Feb 01 '22 05:02 Wansh619

Typically we recommend that you fork the development branch and share a design proposal in the issue or submit a PR.

See issue #92 for a good example that implemented cross-batch memory support.

Feb 07 '22 18:02 owenvallis