Dear authors:
Recently I have looking at your this project and feel interested in it! After I have read your github introduction, the URL is that； https://github.com/jondeaton/BraTS18-Project， and I face some problems. The first question, is that, refer to your introduction like this,
We used skip connections to propagate relevant spatial information from early layers into the later layers. In an effort to examine trade-offs between model accuracy and efficiency, we trained three models each with different skip connections between "down" and "up" blocks. Our first model uses concatenation of "down" and "up" blocks as skip connections and does not have a dropout layer before the final output. Our second model uses element-wise summations of "down" and "up" blocks as opposed to concatenations for skip connections and includes a dropout layer, while the final model uses neither concatenations nor summations.
And I want to know, where is the skip connection in the designed 'UNet' network. Attached the network below. (I could not find the skip connection in this figure)
And the different skip connections, uses concatenation or uses element-wise summations , how could I understand them?
Another question, is that the github introduction part "BraTS Data Loader". Firstly, is the related codes located './BraTS/' ,
And the BraTs data including BraTS2015,BraTS2017,BraTS2018, should I download them at first? Or I just create the BraTS dataset directory like below:
BraTS
├── BraTS15
│ ├── training
│ └── validation
├── BraTS17
│ ├── training
│ └── validation
└── BraTS18
└── training
And I import BraTS package, and use the related interface, and the software could help us to download the data?
And at last, I find a download .sh file,
And I could only see the "brats_2017_train","brats_2017_validation","brats_2018_train", maybe lake the "2015" for train and validation, and '2018' for train?
Thanks a lot! I am looking forward to your reply Sincerely!
Dear authors: Recently I have looking at your this project and feel interested in it! After I have read your github introduction, the URL is that； https://github.com/jondeaton/BraTS18-Project， and I face some problems. The first question, is that, refer to your introduction like this,
We used skip connections to propagate relevant spatial information from early layers into the later layers. In an effort to examine trade-offs between model accuracy and efficiency, we trained three models each with different skip connections between "down" and "up" blocks. Our first model uses concatenation of "down" and "up" blocks as skip connections and does not have a dropout layer before the final output. Our second model uses element-wise summations of "down" and "up" blocks as opposed to concatenations for skip connections and includes a dropout layer, while the final model uses neither concatenations nor summations.
And I want to know, where is the skip connection in the designed 'UNet' network. Attached the network below. (I could not find the skip connection in this figure)
https://user-images.githubusercontent.com/15920014/42143180-b9784310-7d68-11e8-850b-cecf6a3a9175.png
And the different skip connections, uses concatenation or uses element-wise summations , how could I understand them?
Another question, is that the github introduction part "BraTS Data Loader". Firstly, is the related codes located './BraTS/' , And the BraTs data including BraTS2015,BraTS2017,BraTS2018, should I download them at first? Or I just create the BraTS dataset directory like below: BraTS ├── BraTS15 │ ├── training │ └── validation ├── BraTS17 │ ├── training │ └── validation └── BraTS18 └── training
And I import BraTS package, and use the related interface, and the software could help us to download the data? And at last, I find a download .sh file,
!/usr/bin/env bash
brats_2017_train="https://www.cbica.upenn.edu/sbia/Spyridon.Bakas/MICCAI_BraTS/2017/postChallenge/MICCAI_BraTS17_Data_Training_for_CBackes.zip"
brats_2017_validation="https://www.cbica.upenn.edu/sbia/Spyridon.Bakas/MICCAI_BraTS/2017/MICCAI_BraTS17_Data_Validation.zip"
brats_2018_train="https://www.cbica.upenn.edu/sbia/Spyridon.Bakas/MICCAI_BraTS/2018/MICCAI_BraTS_2018_Data_Training.zip"
wget "$brats_2017_train" & wget "$brats_2017_validation" & wget "$brats_2018_train" &
And I could only see the "brats_2017_train","brats_2017_validation","brats_2018_train", maybe lake the "2015" for train and validation, and '2018' for train?
Thanks a lot! I am looking forward to your reply Sincerely!
Best regards.