lack file named val_2D

[cebain]

Hello, I wonder if the folder lacks a file called val_2D, in which there is a function called test_single_volume used to perform validation.

[ziyangwang007]

Hi, could you send me a screenshot of your output/bug/error?

[cebain]

Hi，I can't find these files.

[ziyangwang007]

Thanks for your message, I have fixed it, please kindly check.

[cebain]

hello，thanks for your revision very much。 I still want to ask a question。 We can see that the h5 file includes both image and label data， and when we train，for unlabeld data， we just extract volume batch which is the image data。 However， if I don't have labels of all data，I just have labels for some of the data, which really needs smei-supervise, then how should I write
the h5 file, or I just read image for 2D pictures?

[ziyangwang007]

Hi, I am not sure whether I understand your question, but I assume you are asking about the ratio of labeled/unlabeled training set. Just simply set the --labeled_num XXX (for example 25) when you train the model.

[cebain]

不好意思啊，我英语不太好，我直接用中文哈，就是 现在的acdc数据集，每一个文件都有image和label，然后每个slice的h5文件是包含image和label的，然后在半监督训练的时候，对于设定无标签的序号，在取数据的时候，只取出image。 但是一般情况下，我本身有N个2D图片，M个标签（M<N), 那么对于N-M个无标签的数据，h5文件写入的时候只要写入image 就可以了是吧， 这样的话，读取的时候也不能直接用two_batch_sampler ，需要重新写一个读取策略 | | 崔灿灿 | | ***@***.*** | ---- Replied Message ---- | From | ***@***.***> | | Date | 2/10/2023 23:12 | | To | ***@***.***> | | Cc | ***@***.***> , ***@***.***> | | Subject | Re: [ziyangwang007/CV-SSL-MIS] lack file named val_2D (Issue #2) | Thanks for your message, I have fixed it, please kindly check. — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you authored the thread.Message ID: ***@***.***>

[ziyangwang007]

是的，你说的很对。如果是你自己的数据集（比如M个标签 N个图片），需要修改一下数据读取部分代码。

[cebain]

好滴，谢谢！我修改试试！ | | 崔灿灿 | | ***@***.*** | ---- Replied Message ---- | From | ***@***.***> | | Date | 2/16/2023 01:27 | | To | ***@***.***> | | Cc | ***@***.***> , ***@***.***> | | Subject | Re: [ziyangwang007/CV-SSL-MIS] lack file named val_2D (Issue #2) | 是的，你说的很对。如果是你自己的数据集（比如M个标签 N个图片），需要修改一下数据读取部分代码。 — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you authored the thread.Message ID: ***@***.***>

[cebain]

嗯嗯，我更换了数据读取方式，可以进行2D图片的多类别分割，不过出现了一个神奇的事情，就是在测试的时候，vit竟然可以加载unet的权重且不报错，虽然这样分割出来的结果是这样 哈哈，还需要再探究一下。 在 2023-02-16 01:27:35，"ziyangwang007" ***@***.***> 写道： 是的，你说的很对。如果是你自己的数据集（比如M个标签 N个图片），需要修改一下数据读取部分代码。 — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you authored the thread.Message ID: ***@***.***>