How to add a review

The process for adding reviews is git-centric. Basically, you just need to add a file to the repo and make a pull request. Let’s go into the details :

  1. Send an e-mail to Pierre-Marc Jodoin asking him to include you as a member of the github.com/vitalab organization.
  2. Clone the vitalab.github.io repo on your computer.
  3. Set up the pre-commit hooks by executing the utils/setup_hooks.sh script.
  4. Determine the category in which you will add your post. Categories are managed using folders :

    _posts/           # A post added here will have no category
    course/
        _posts/       # A post added here will be in the "course" category
    machine-learning/
        _posts/       # Same thing for the "machine-learning" category
    ...               # There are other categories, you can add one too.
    
  5. Create a file YYYY-MM-DD-title-of-your-review.markdown and put it in right folder (see above).
    It is important that you respect this format : date at the beginning and no spaces. Else the page won’t build properly. Here is an example of a valid name : 2017-01-31-going-deeper-with-convolutions.markdown.
  6. Use the review template file in the templates as a starting point and do your review.

    A minimal working review example may look like:

    ---
    layout: review
    title: U-Net Convolutional Networks for Biomedical Image Segmentation
    tags: deep-learning CNN segmentation medical essentials
    cite:
        authors: "O. Ronneberger, P. Fischer, T. Brox"
        title:   "U-Net: Convolutional Networks for Biomedical Image Segmentation"
        venue:   "Proceedings of MICCAI 2015, p.234-241"
    pdf: "https://arxiv.org/pdf/1505.04597.pdf"
    ---
    
    # Introduction
    
    Famous 2D image segmentation CNN made of a series of convolutions and
    deconvolutions. The convolution feature maps are connected to the deconv maps of
    the same size. The network was tested on the 2 class 2D ISBI cell segmentation
    [dataset](http://www.codesolorzano.com/Challenges/CTC/Welcome.html).
    Used the crossentropy loss and a lot of data augmentation.
    
    The network architecture:
    ![](/article/images/MyReview/UNetArchitecture.png)
    
    A U-Net is based on Fully Convolutional Networks (FCNNs)[^1].
    
    The loss used is a cross-entropy:
    $$ E = \sum_{x \in \Omega} w(\bold{x}) \log (p_{l(\bold{x})}(\bold{x})) $$
    
    The U-Net architecture is used by many authors, and has been re-visited in
    many reviews, such as in [this one](https://vitalab.github.io/article/2019/05/02/MRIPulseSeqGANSynthesis.html).
    
    # References
    
    [^1]: Jonathan Long, Evan Shelhamer, and Trevor Darrell. Fully convolutional
          networks for semantic segmentation (2014). arXiv:1411.4038.
    

    The list of available tags can be modified to include new tags. The current ones are:


3D CNN DBM GAN
GMM GNN GOFAI KDE
LSTM MRI NAS PCA
RBM RNN VAE action-recognition
active-learning adversarial attention attention-maps
auto-ml autoencoder benchmarking blog
bounding-boxes brain caption cardiac
classification codebook course ct-scan
data-augmentation dataset deep-learning denoising
detection dimensionality-reduction domain-adaptation essentials
face-detection few-shot-learning genetic-algorithm graph-cut
hyperspectral imitation-learning inpainting k-means
layers localization machine-learning medical
meta-learning motion-detection multi-agent multi-task
multi-task-learning multiple-sclerosis network-compression network-pruning
neural-network one-shot-segmentation optimization pedestrian-detection
point-cloud pruning-acceleration reinforcement remote-sensing
segmentation sequence shape-analysis siamese
surveillance survey synthesis tractography
tractometry traffic transfer-learning unsupervised
video-analysis weakly-supervised white-matter


NOTE: the essentials tags is used for any paper considered as being a “must-read”.

For further information about Jekyll’s syntax, visit the documentation page. However, note that Jekyll’s syntax may change in newer versions, and the site’s version is freezed. Hence, although Liquid tags could be used for links, for example, plain old links are used to avoid issues building the site.

You can preview your post while you write it ; see the next section about this.\

  1. Make a new branch, commit your file and push your branch.
  2. Create a pull request on the repo’s github page.
  3. Add reviewers: everyone that you think are knowledgeable about the subject or simply would be interested in your review.
  4. When every reviewer approved your branch, merge your branch and delete it.

How to preview your post locally

This site is built around Jekyll. Jekyll takes all the markdown files and generates a static html website.

  1. Install Ruby using rbenv. Don’t use apt-get since its version of Ruby is too old.
  2. Install Jekyll by running : gem install bundler jekyll.
  3. Go where you cloned the VITAL literature review repository and run : bundle install. This will install the dependencies for our Jekyll site.
  4. Run a local webserver using : bundle exec jekyll serve.
  5. Access the site locally at : http://127.0.0.1:4000/

Note that the site is automatically rebuilt when a file has been modified.