Hyper Parameter for Resnet18 Model

[ganguli9082]

Hi,

Just curious about the parameters that are used for training the model for polyp type on the 7000um(subsample 224) patches and the grade classification that was done on the 800um patches for both the subsample of 224x224 and no subsample. I started out using the default settings in the train.py but can't seem to get close to the results described in the paper.

Thank you for your time.
Alex Ganguli

Preprocess: HE
Apply Transformations: Training images= True,
Testing images = False

learning rate: 0.01
batch size: 256
num workers: 8
decay factor: 0.1
step size: 20

[carloalbertobarbano]

Hello, sorry for the delay in the response.

So, the are two main tasks:

1. Type classification on 7000µm patches -> here we subsample to 224
2. Grade prediction on 800µm patches -> here we do not subsample the input image

In any case the preprocessing is RGB. Can you try replicating task 2 with one of these hyperparameters settings: https://wandb.ai/eidos/UnitoPath-v1/reports/Grade-predictions--VmlldzoxMzY4NzI5 that should get you around 80% BA for grade prediction (you can click on a single run then go to overview to get the full list of arguments)

[ganguli9082]

Great!! Thanks for getting back to me. I’ll let you know how it goes, thank you. Cheers, Alex Ganguli

…

On Wed, Dec 22, 2021 at 5:21 AM Carlo Alberto Barbano < ***@***.***> wrote: Hello, sorry for the delay in the response. So, the are two main tasks: 1. Type classification on 7000µm patches -> here we subsample to 224 2. Grade prediction on 800µm patches -> here we do not subsample the input image In any case the preprocessing is RGB. Can you try replicating task 2 with one of these hyperparameters settings: https://wandb.ai/eidos/UnitoPath-v1/reports/Grade-predictions--VmlldzoxMzY4NzI5 that should get you around 80% BA for grade prediction — Reply to this email directly, view it on GitHub <#1 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AMC5G6M5GQA32HXCVKMGDO3USHGGHANCNFSM5KOUN4FQ> . Triage notifications on the go with GitHub Mobile for iOS <https://apps.apple.com/app/apple-store/id1477376905?ct=notification-email&mt=8&pt=524675> or Android <https://play.google.com/store/apps/details?id=com.github.android&referrer=utm_campaign%3Dnotification-email%26utm_medium%3Demail%26utm_source%3Dgithub>. You are receiving this because you authored the thread.Message ID: ***@***.***>