Data

Tools

Indexes

Applications

Experiments

Open Source Science

multiscale-adversarial-attention-gates

Code for the paper: Valvano G. et al. (2021), Learning to Segment from Scribbles using Multi-scale Adversarial Attention Gates

https://github.com/gvalvano/multiscale-adversarial-attention-gates

Science Score: 67.0%

This score indicates how likely this project is to be science-related based on various indicators:

  • ✓
    CITATION.cff file
    Found CITATION.cff file
  • ✓
    codemeta.json file
    Found codemeta.json file
  • ✓
    .zenodo.json file
    Found .zenodo.json file
  • ✓
    DOI references
    Found 1 DOI reference(s) in README
  • ✓
    Academic publication links
    Links to: arxiv.org
  • ○
    Academic email domains
  • ○
    Institutional organization owner
  • ○
    JOSS paper metadata
  • ○
    Scientific vocabulary similarity
    Low similarity (14.5%) to scientific vocabulary

Keywords

attention cnn deep-learning gan generative-adversarial-network segmentation semantic-segmentation weakly-supervised-learning
Last synced: 6 months ago · JSON representation ·

Repository

Code for the paper: Valvano G. et al. (2021), Learning to Segment from Scribbles using Multi-scale Adversarial Attention Gates

Basic Info
  • Host: GitHub
  • Owner: gvalvano
  • License: apache-2.0
  • Language: Python
  • Default Branch: master
  • Homepage:
  • Size: 2.39 MB
Statistics
  • Stars: 43
  • Watchers: 3
  • Forks: 9
  • Open Issues: 2
  • Releases: 0
Topics
attention cnn deep-learning gan generative-adversarial-network segmentation semantic-segmentation weakly-supervised-learning
Created over 5 years ago · Last pushed about 4 years ago
Metadata Files
Readme License Citation

README.md

Learning to Segment from Scribbles using Multi-scale Adversarial Attention Gates

Code for the paper:

Valvano, Gabriele, Andrea Leo, and Sotirios A. Tsaftaris. "Learning to Segment from Scribbles using Multi-scale Adversarial Attention Gates." IEEE Transactions on Medical Imaging (2021).

The official project page is here.
An online version of the paper can be found here.

Citation:

@ARTICLE{9389796, author={Valvano, Gabriele and Leo, Andrea and Tsaftaris, Sotirios A.}, journal={IEEE Transactions on Medical Imaging}, title={Learning to Segment from Scribbles using Multi-scale Adversarial Attention Gates}, year={2021}, volume={}, number={}, pages={1-1}, doi={10.1109/TMI.2021.3069634} }

mscale_aags

Notes:

You can find the entire tensorflow model inside expriments/acdc/model.py. This file contains the main class that is used to train on the ACDC dataset. Please, refer to the class method define_model() to see how to correctly build the CNN architecture. The structure of the segmentor and the discriminator alone can be found under the folder architectures.

Once you download the ACDC dataset and the scribble annotations, you can pre-process it using the code in the file data_interface/utils_acdc/prepare_dataset.py. You can also train with custom datasets, but you must adhere to the template required by data_interface/interfaces/dataset_wrapper.py, which assumes the access to the dataset is through a tensorflow dataset iterator.

Once preprocessed the data, you can start the training running the command:
python -m train --RUN_ID="${run_id}"_${perc}_${split} --n_epochs=450 --CUDA_VISIBLE_DEVICE=${CUDA_VD} --data_path=${dpath} --experiment="${path}" --dataset_name=${dset_name} --verbose=True --results_dir=${res_dir} --n_sup_vols=${perc} --split_number=${split}
This will train the model and do a final test on the ACDC dataset. If you also want to test the results using the challenge server, after running the above command, you must run:
python -m test_on_acdc_test_set --RUN_ID="${run_id}"_${perc}_${split} --CUDA_VISIBLE_DEVICE=${CUDA_VD} --data_path=${dpath} --experiment="${path}" --dataset_name=${dset_name} --verbose=False --n_sup_vols=${perc} --split_number=${split}
and then submit the results as explained here.

Refer to the file run.sh for a complete example.

Requirements

This code was implemented using TensorFlow 1.14. We tested it on a TITAN Xp GPU, and on a GeForce GTX 1080, using CUDA 8.0, 9.0 and 10.2.

Owner

  • Name: Gabriele Valvano
  • Login: gvalvano
  • Kind: user

AI Specialist working on Computer Vision and Generative Models

Citation (CITATION.cff) 

cff-version: 1.2.0
message: "If you use this software, please cite it as below."
authors:
- family-names: "Valvano"
  given-names: "Gabriele"
- family-names: "Leo"
  given-names: "Andrea"
- family-names: "Tsaftaris"
  given-names: "Sotirios A."
title: "Learning to Segment from Scribbles using Multi-scale Adversarial Attention Gates"
version: 1.0.0
doi: 10.1109/TMI.2021.3069634
date-released: 2021
url: "https://github.com/gvalvano/multiscale-adversarial-attention-gates"

GitHub Events

Total
Last Year