Training on other dataset

[rik030]

Can I run this architecture on MyoPS 2020 dataset which have 6 class segmentation?

[mdciri]

Dear Rishav, Yes you can. On the Readme.md file there is written how: python ./main.py '-nc' or '--num_classes': number of classes, default 4 Best regards, Marco

…

On Sun, Feb 27, 2022, 07:15 Rishav Rajak ***@***.***> wrote: Can I run this architecture on MyoPS 2020 dataset which have 6 class segmentation? — Reply to this email directly, view it on GitHub <#9>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AOJQHCEWHSXMOODLEVULPEDU5G6ORANCNFSM5POMNSVA> . 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 are subscribed to this thread.Message ID: ***@***.***>

[rik030]

Hello Sir,
In the Brats20 dataset, the modalities are 240x240x155 for every patient but in MyoPS dataset, shapes are different for different patients like patient 1 has 192x168x5 for each modality while patient 2 has 255x179x3 for each modality. Here we have only 3 modalities (bssfp, T2, and LGE). How to tackle this situation?

[mdciri]

You should fix your dataset first in order to have images/volumes with same shape (zero padding may help). Then if your dataset images/volumes have 3 channels you should check, and in case edit, the input layer of the generator and discriminator. Best regards, Marco Domenico Cirillo

…

On Sun, Feb 27, 2022, 12:12 Rishav Rajak ***@***.***> wrote: Hello Sir, In the Brats20 dataset, the modalities are 240x240x155 for every patient but in MyoPS dataset, shapes are different for different patients like patient 1 has 192x168x5 for each modality while patient 2 has 255x179x3 for each modality. Here we have only 3 modalities (bssfp, T2, and LGE). How to tackle this situation? — Reply to this email directly, view it on GitHub <#9 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AOJQHCA73V3N22S7PTBJ4BDU5IBJ3ANCNFSM5POMNSVA> . 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 commented.Message ID: ***@***.***>

[rik030]

Hello Sir,
From zero padding, I can change the first two dimensions to get the same shape, but for the number of slices(which varies from 3 to 5), how can I make it the same?

[mdciri]

I don't know your dataset. But what you can do is or adding 0s slices to those volumes with 3, or remove 2 to those with 5. All volumes must have same shape. Another thing you can do is training a cycle gan to go from volumes with 3 slices to volumes with 5. Anyway, these solving problems do no concern the vox2vox model. Best regards, Marco Domenico Cirillo

…

On Sun, Feb 27, 2022, 14:46 Rishav Rajak ***@***.***> wrote: Hello Sir, From zero padding, I can change the first two dimensions to get the same shape, but for the number of slices(which varies from 3 to 5), how can I make it the same? — Reply to this email directly, view it on GitHub <#9 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AOJQHCHZKHPGJA7JQQX6YD3U5ITKLANCNFSM5POMNSVA> . 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 commented.Message ID: ***@***.***>

[rik030]

Hello Sir,
Wouldn't adding 0s slices to those volumes with 3 or 4, affect the model training, as we are assigning class 0 to the background in the ground truth as well.