Closed lyc-lang closed 4 years ago
@lyc-lang This field is more than 1.5GB, so I think I can't send you by email. I'm tried to diagnose your problem. But the link seems to work good. I just downloaded it just few minutes ago. Can you try it again?
Many thanks,I have worked it out in another way.
As you said that you have implemented a cross validation with Corresponding coefficient k = 6
, but I found in your dataset.py code self.train_data = [item for sub_block in blocks for item in sub_block] + \filenames[-np.mod(len(filenames), num_cross_vals):]
,In this case, train_data overlaps with val_data and test_data when the data set is partitioned as you do. Any questions here?
@lyc-lang
DCNN model is trained several times and each time the model uses different separations of the train, validation, and test. Therefore, they are not overlapped. The variable, idx_cross_val
plays an essential role to control different validation and test.
I am sorry to disturb you again.
I know what you mean, however, when I took 10 files to test result, for example,I take k =5,Dataset('brain01', 5, 0)
.the outcome shows the result that test_data = [0, 1], val_data = [2, 3], train_data =[4,5,6,7,8,9,0,1,2,3,4,5,6,7,8,9](each number represents a file name).Then I make a change by self.train_data = [item for sub_block in blocks for item in sub_block]
,and get result as test_data = [0, 1], val_data = [2, 3], train_data =[4,5,6,7,8,9], So is there a little bit of a flaw in the code here?
@lyc-lang
Yes, you are right. The +filenames[-np.mod(len(filenames), num_cross_val):]
wants to use the remaining data if the number of data can't divide by the number of blocks correctly. In your experiments, If you have 11 files and 5 blocks, then simply ignoring the above code can't use the last data file.
The bug if fixed as the following:
Thank you for your careful testing.
In that case, the outcome of data separation is right. Thank you for reminding me!
The link here cannot be downloaded to the trained VGG16 model weights. Could you please send me a copy? 79808121@qq.com Thank you
I'm sorry!I can't send u now,maybe I could give u a copy in the next few weeks if I can have access to the internet.
---Original--- From: "zichao999"<notifications@github.com> Date: Sat, Feb 29, 2020 22:40 PM To: "ChengBinJin/MRI-to-CT-DCNN-TensorFlow"<MRI-to-CT-DCNN-TensorFlow@noreply.github.com>; Cc: "lyc-lang"<mirror_whut@qq.com>;"Mention"<mention@noreply.github.com>; Subject: Re: [ChengBinJin/MRI-to-CT-DCNN-TensorFlow] model (#2)
The link here cannot be downloaded to the trained VGG16 model weights. Could you please send me a copy? 79808121@qq.com Thank you
— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub, or unsubscribe.
@zichao999 Hi, you can find the pretrained VGG model from the following link (https://yunpan.360.cn/surl_yxLDnu6QqjQ (password:3ouy)
@lyc-lang The link here cannot be downloaded to the trained VGG16 model weights. Could you please send me a copy? 3517162165@qq.com Thank you!
The link here cannot be downloaded to the trained VGG16 model weights. Could you please send me a copy?
mirror_whut@qq.com
Thank you