Hi I have a questions about experiments

[Ha-Jungmin]

First, Thanks for good research and i'm interested in skin cancer lession classification task.
So i want try reproduce your experiments but i only got 94% accuracy on HAM10000 dataset.
I execute 01_Skin_Distinction.ipynb, 02_Aug_img.ipynb, 03_Svd_Blend.ipynb to make datasets and i make 54292 images to train and 828 images to validation.
And i use FixCaps_HAM_24.ipynb to train and validate FixCaps model but I got 94% accuracy.
Is there any train methods or I did some mistakes?

And another question is about random seed.
I tried to fix seeds to use

np.random.seed(10)
torch.manual_seed(10)
torch.cuda.manual_seed(10)

these three codes but i can't fix result.
Is there any other methods?

Thank you

[Woodman718]

I'm glad you're interested in my research.
You can try my newly uploaded code(FixCaps_HAM-29.ipynb).
There are two factors that extremely influence the performance.
One is the preprocessing of the data, the relevant code is derived from[32](C. Zhao etal, "Dermoscopy Image Classification Based on StyleGAN and DenseNet2"). The data for the test set were randomly drawn from a specially selected set of 4501 samples(like the "lesion_id_HAM_0000005").
Another major factor is the random initialization of the weights. David Picard's work (" Torch.manual_seed is all you need On the Influence of Random Seeds") should help you.
And it is undeniable that the final choice of hyperparameters has a certain impact on the final performance of the model.
The robustness of the model does need to be improved, and recently I've been experimenting with new ideas. However, due to the difficulties of various trifles, the progress of the research is not particularly smooth. If I have further results, I will publish them on github or in a new paper.

[Woodman718]

We can communicate by email ([email protected]).

[Woodman718]

​I have identified the factors that cause model instability: FractionalMax pooling, Squash function, random seeds, BatchSize, etc., and have overcome these flaws. I will post the code when the paper on FixCaps-V2 is published.

[Ha-Jungmin]

I make a sense your explanation Very thank you for your careful explanation 2022년 11월 30일 (수) 오후 8:33, Woodman718 ***@***.***>님이 작성:

…

​I have identified the factors that cause model instability: FractionalMax pooling, Squash function, random seeds, BatchSize, etc., and have overcome these flaws. I will post the code when the paper on FixCaps-V2 is published. — Reply to this email directly, view it on GitHub <#1 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ASHBKDUAWI7MS5HEGTZEFKLWK43QRANCNFSM6AAAAAARXZISJM> . You are receiving this because you authored the thread.Message ID: ***@***.***>

--

[farisbasha]

Hey can i get tensorflow version of fixcaps model?

[Woodman718]

Sorry, we currently have no plans to provide a TF version.