about the data format

[qinhaifangpku]

hi~ I am very appreciate your post this work! it is very helpful! I wonder the datasets format you provide for train and test. would you please give me a example? is it hdf5? I noticed that you read the image. but the volume data is 3D. would you please give me some example about the data format? thank you in advance!

[faustomilletari]

Hello, the data format is mhd. Referring to the prostate segmentation challenge, you can download all the images and place them into two directories (train and test) and start from there. to reproduce the results of my paper you will need to pre-process them, even though you will get pretty close results also using non pre-processed volumes.

in summary just specify in the config the directories where your mhd files lie.

Fausto

On Jan 16, 2017, at 9:04 AM, qinhaifangpku notifications@github.com wrote:

hi~ I am very appreciate your post this work! it is very helpful! I wonder the datasets format you provide for train and test. would you please give me a example? is it hdf5? I noticed that you read the image. but the volume data is 3D. would you please give me some example about the data format? thank you in advance!

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/faustomilletari/VNet/issues/11, or mute the thread https://github.com/notifications/unsubscribe-auth/AMtsvvyb_ajmlC8hylP_js0OtFVvPkldks5rS3jrgaJpZM4Lklr3.

[qinhaifangpku]

thank you！are gray image and label image both mhd format？

在 2017年1月17日，00:35，Fausto Milletari notifications@github.com 写道：

Hello, the data format is mhd. Referring to the prostate segmentation challenge, you can download all the images and place them into two directories (train and test) and start from there. to reproduce the results of my paper you will need to pre-process them, even though you will get pretty close results also using non pre-processed volumes.

in summary just specify in the config the directories where your mhd files lie.

Fausto

On Jan 16, 2017, at 9:04 AM, qinhaifangpku notifications@github.com wrote:

hi~ I am very appreciate your post this work! it is very helpful! I wonder the datasets format you provide for train and test. would you please give me a example? is it hdf5? I noticed that you read the image. but the volume data is 3D. would you please give me some example about the data format? thank you in advance!

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/faustomilletari/VNet/issues/11, or mute the thread https://github.com/notifications/unsubscribe-auth/AMtsvvyb_ajmlC8hylP_js0OtFVvPkldks5rS3jrgaJpZM4Lklr3.

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub, or mute the thread.

[faustomilletari]

Yes, I recall that the annotations had a "_segmentation" in their filenames...

Fausto Milletarì Sent from my iPhone

On 16 Jan 2017, at 21:00, qinhaifangpku notifications@github.com wrote:

thank you！are gray image and label image both mhd format？

在 2017年1月17日，00:35，Fausto Milletari notifications@github.com 写道：

Hello, the data format is mhd. Referring to the prostate segmentation challenge, you can download all the images and place them into two directories (train and test) and start from there. to reproduce the results of my paper you will need to pre-process them, even though you will get pretty close results also using non pre-processed volumes.

in summary just specify in the config the directories where your mhd files lie.

Fausto

On Jan 16, 2017, at 9:04 AM, qinhaifangpku notifications@github.com wrote:

hi~ I am very appreciate your post this work! it is very helpful! I wonder the datasets format you provide for train and test. would you please give me a example? is it hdf5? I noticed that you read the image. but the volume data is 3D. would you please give me some example about the data format? thank you in advance!

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/faustomilletari/VNet/issues/11, or mute the thread https://github.com/notifications/unsubscribe-auth/AMtsvvyb_ajmlC8hylP_js0OtFVvPkldks5rS3jrgaJpZM4Lklr3.

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub, or mute the thread.

— You are receiving this because you commented. Reply to this email directly, view it on GitHub, or mute the thread.

[toki2015]

Hi

Could you explain the meaning of dimensions of "data" in "test_noPooling_ResNet_cinque.prototxt"

input: "data" input_shape{dim: 1 dim: 1 dim: 128 dim: 128 dim: 64}

I understood [dim: 128 dim: 128 dim: 64] means the data size of your 3D MR data what is meaning about first dim: 1 and second dim: 1 ?

Thank you in advance

[faustomilletari]

First dimension is the batch size. Second the number of channels

Fausto Milletarì Sent from my iPhone

On 19 Jan 2017, at 04:02, toki2015 notifications@github.com wrote:

Hi

Could you explain the meaning of dimensions of "data" in "test_noPooling_ResNet_cinque.prototxt"

input: "data" input_shape{dim: 1 dim: 1 dim: 128 dim: 128 dim: 64}

I understood [dim: 128 dim: 128 dim: 64] means the data size of your 3D MR data what is meaning about first dim: 1 and second dim: 1 ?

Thank you in advance

— You are receiving this because you commented. Reply to this email directly, view it on GitHub, or mute the thread.

[toki2015]

Thank you very much! Batch size is related to the number of image cases used for testing or training each time, right ? I am considering to use VNET for multiple target segmentations, that each ground truth has more than 2 labels. Is it possible to do it by slightly changing some part in your VNET ? In this case, should we change the batch size (I think it was 2 in you current setting ) ?

[faustomilletari]

On Jan 19, 2017, at 7:51 PM, toki2015 notifications@github.com wrote:

Thank you very much! Batch size is related to the number of image cases used for testing or training each time, right ?

it’s more, the number of volumes fed to the network at each iteration of optimization. see minibatch online.

I am considering to use VNET for multiple target segmentations, that each ground truth has more than 2 labels. Is it possible to do it by slightly changing some part in your VNET ?

I cannot answer this question at this time. I guess you could invest some time in some changes, but this depends on your familiarity with caffe framework.

In this case, should we change the batch size (I think it was 2 in you current setting ) ?

the batch size is not directly affected by the number of labels.

— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/faustomilletari/VNet/issues/11#issuecomment-273945456, or mute the thread https://github.com/notifications/unsubscribe-auth/AMtsvgqLFTEZj_RBjsDLdeyzTkB1Uh7Oks5rUAT1gaJpZM4Lklr3.

[sagarhukkire]

Hi Fausto

Just to clarify I got ,mhd files for labels and images ..as you said you kept files in Train and Test folder. So to train network I just need to place labels and images in Train folder or I need to keep separate? Does there any provision where I can just train then I test network? Thanks

[faustomilletari]

currently you need to have a directory

• — train
  • filename1.mhd
  • filename1.raw
  • filename1_segmentation.mhd
  • filename1_segmentation.raw
  • ….
    • ….

and a directory

• — test
  • filenametest1.mhd
  • filenametest1.raw
  • ….
    • ….

the test directory can also be empty!

On Feb 6, 2017, at 10:01 AM, sagarax009 notifications@github.com wrote:

Hi Fausto

Just to clarify I got ,mhd files for labels and images ..as you said you kept files in Train and Test folder. So to train network I just need to place labels and images in Train folder or I need to keep separate? Does there any provision where I can just train then I test network? Thanks

— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/faustomilletari/VNet/issues/11#issuecomment-277708043, or mute the thread https://github.com/notifications/unsubscribe-auth/AMtsvjSLNWyKkLxLhxiPWVWqjhQPqkMPks5rZzW5gaJpZM4Lklr3.

[sagarhukkire]

ok thanks..

[sagarhukkire]

.raw is another file format or you are saving the same file in two format from MITK??

[faustomilletari]

Hello,

mhd files have an header that is .mhd and a “data file” that is the .raw.

The presence of .raw files is part of the mhd format.

to handle other formats you have to change the data manager class at line

https://github.com/faustomilletari/VNet/blob/master/DataManager.py#L24 https://github.com/faustomilletari/VNet/blob/master/DataManager.py#L24

to change the fact that segmentation volumes should terminate with _segmentation you change line

https://github.com/faustomilletari/VNet/blob/master/DataManager.py#L32 https://github.com/faustomilletari/VNet/blob/master/DataManager.py#L32

regards,

Fausto Milletari

On Feb 6, 2017, at 10:51 AM, sagarax009 notifications@github.com wrote:

.raw is another file format or you are saving the same file in two format from MITK??

— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/faustomilletari/VNet/issues/11#issuecomment-277723443, or mute the thread https://github.com/notifications/unsubscribe-auth/AMtsvjpniRDB0fNRlBeNoPGrbFfQKmybks5rZ0F1gaJpZM4Lklr3.

[sagarhukkire]

Thanks man ..since this is Medical image which are complete new so all this questions but hope I will now manage now... Thanks alot

[faustomilletari]

No problems! good luck with your project!

On Feb 6, 2017, at 10:57 AM, sagarax009 notifications@github.com wrote:

Thanks man ..since this is Medical image which are complete new so all this questions but hope I will now manage now... Thanks alot

— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/faustomilletari/VNet/issues/11#issuecomment-277725379, or mute the thread https://github.com/notifications/unsubscribe-auth/AMtsvhaSYGTgfkkm0h2VbspZ8JwrUDklks5rZ0LYgaJpZM4Lklr3.

[Jorisfournel]

Hi,

I have .hdr and .img files could it work directly ? I look for a way to convert them to mhd and raw but to this point I didn't succeed. Could I change something in the datamanager file so that it could work in hdr ?

Thanks in advance and for your work !

[faustomilletari]

Hello,

I’m unfamiliar with HDR, but if it is an ITK file format you will be good to go with what is already coded. You can also, of course, write your own data manager class that implements what you need.

Fausto

On Oct 18, 2017, at 5:00 AM, Jorisfournel notifications@github.com wrote:

Hi,

I have .hdr and .img files could it work directly ? I look for a way to convert them to mhd and raw but to this point I didn't succeed. Could I change something in the datamanager file so that it could work in hdr ?

Thanks in advance and for your work !

— You are receiving this because you modified the open/close state. Reply to this email directly, view it on GitHub https://github.com/faustomilletari/VNet/issues/11#issuecomment-337567820, or mute the thread https://github.com/notifications/unsubscribe-auth/AMtsvguTP7ZPX--g0RgI1Ujxjcq6G2IDks5steh3gaJpZM4Lklr3.