about retraining cnn explainer

[BobBogart]

Hello,it's really happy using cnn explainer.Yet I have a question,how could we use it to train a cnn for things like sword with around 200 kinds?

[xiaohk]

Hello @BobBogart,

I am happy to hear that you like CNN Explainer!

To use CNN Explainer with your own model and dataset, you can check out these related issues: #2, #8, #15. Some functions you want to look into are listed in this comment.

If you want to train Tiny-VGG, the model used in CNN Explainer, on the sword dataset, you can check out issue #14. They have successfully trained Tiny-VGG on their plant dataset and visualize the CNN with CNN Explainer. Since 200 classes are a lot, perhaps you can try to choose 10 most distinct classes as a first step. Later you can modify the CNN Explainer UI code to have more output neurons (you can keep the number of neurons in hidden layers as 10).

I will close the issues for now. Feel free to reach out if you need further help :)

[BobBogart]

Thank you very much!

…

---Original--- From: "Jay ***@***.***> Date: Sun, Jun 13, 2021 18:59 PM To: ***@***.***>; Cc: ***@***.******@***.***>; Subject: Re: [poloclub/cnn-explainer] about retraining cnn explainer (#31) Hello @BobBogart, I am happy to hear that you like CNN Explainer! To use CNN Explainer with your own model and dataset, you can check out these related issues: #2, #8, #15. Some functions you want to look into are listed in this comment. If you want to train Tiny-VGG on the sword dataset, you can check out issue #14. They have successfully trained Tiny-VGG on their plant dataset and visualize the CNN with CNN Explainer. Since 200 classes are a lot, perhaps you can try to choose 10 most distinct classes as a first step. Later you can modify the CNN Explainer UI code to have more output neurons (you can keep the number of neurons in hidden layers as 10). I will close the issues for now. Feel free to reach out if you need further help :) — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub, or unsubscribe.