search

KerrFitzgerald / Polyp_FCB-SwinV2Transformer

Code Implementation of the FCB-SwinV2 Transformer
MIT License
4 stars 1 forks source link

Model implementation #1

Closed long-nguyen12 closed 7 months ago

long-nguyen12 commented 7 months ago

Thanks for your sharing about you great work. In your paper, you compared your model with DUCKNet which is original in TF/Keras. Did you reprocedure DUCKNet in Pytorch? If yes, can you share your DUCKNet implementation in Pytorch? I want to compare with both your models for research purpose. Thanks

KerrFitzgerald commented 7 months ago

Thank you for your kind comments.

In the work for the paper I used the pre-trained TF model from the original DUCKNet authors as I needed to determine the impact of different metric averaging methodologies on the exact same test data partitions that were used by the authors.

I did try to recreate the DUCKNet model using PyTorch but didn't have time to fully resolve some minor issues that existed in my implementation. In the future I would also like to investigate DUCKNet more due to its ability to achieve amazing results without pre-training. I should have some free time later this year so if I do create a fully working PyTorch implementation I will let you know.

long-nguyen12 commented 7 months ago

I understood your comparison scenarios. Thanks a lot