What steps to perform Contrastive Pretraining

[mustansarfiaz]

Hi,

Thank you for updating your code. However, I need your help to perform contrastive pretraining.
Please correct me, If I am wrong;

for the MoNuSAC example, according to my understanding, the steps are as:
1)- Generate cropped patches
2)- Run train_inc_dec_size_weight_balance.py
3)-Run make_pca_model.ipynb

However, In train_inc_dec_size_weight_balance.py, you are using three extra paths, how to generate patches for them?
1)-MoNuSac_cropped_patches/train/ [generated from create_contrastive_patches.py]
2)- MoNuSac_cropped_patches_5P/train/ [?? how do generate these patches and what is meant by 5P??]
3)- MoNuSac_cropped_patches_10P/train/ [?? how do generate these patches and what is meant by 10P??]
4)- MoNuSac_cropped_patches_20P/train/ [?? how do generate these patches and what is meant by 20P??]
5)-MoNuSac_cropped_patches/test/ [generated from create_contrastive_patches.py]

Finally, the config file is missing from Contrastive_pretraining directory.

It would be great if you help us out.

Regards
Mustansar

[Aayushktyagi]

Yes your understanding is right.
I have uploaded the dataset and updated the readme file as well.
You can find the dataset here

Thanks

[mustansarfiaz]

Hi
Thank you for providing the dataset.
Now, can you please provide me the config file [hyperparameters] for Contrastive_pretraining.
Also, it would be good for the users if you also upload the pretrained weights.

Once again,
thank you.

[Aayushktyagi]

Hi,
I have updated pretrained weights and PCA file for ConSep and MoNuSec. I will update config file in coming days.

Thanks
Aayush

[ptrahuang]

Hello, when will the config be uploaded?