Endocv2021-winner
[Paper]
This is the winning solution of the Endocv-2021 grand challange.
Dependencies
pytorch # tested with 1.7 and 1.8
torchvision
tqdm
pandas
numpy
albumentations # for augmentations
torchsummary
segmentation_models_pytorch # for basic segmentaion models
pyra_pytorch # pyra_pytorch.PYRADatasetFromDF is used. But this can be replaced with normal pytorch dataset.
Tri-Unet
Block diagram of Tri-Unet
How to train Tri-Unet and other basic models to DivergentNet?
# To train Tri-unet
python tri_unet.py train \
--num_epochs 2 \
--device_id 0 \
--train_CSVs sample_CSV_files/C1.csv sample_CSV_files/C1.csv \
--val_CSVs sample_CSV_files/C2.csv sample_CSV_files/C3.csv \
--test_CSVs sample_CSV_files/C3.csv \
--out_dir ../temp_data \
--tensorboard_dir ../temp_data
# To train other models, you have to replace tri_unet.py with one of the follwings:
unet_plusplus.py
deeplabv3.py
deeplabv3_plusplus.py
Pretrained checkpoint paths
- https://github.com/vlbthambawita/divergent-net/releases/download/checkpoints_v1/best_checkpoint_Deeplabv3.pth
- https://github.com/vlbthambawita/divergent-net/releases/download/checkpoints_v1/best_checkpoint_Depplabv3_plusplus.pth
- https://github.com/vlbthambawita/divergent-net/releases/download/checkpoints_v1/best_checkpoint_FPN.pth
- https://github.com/vlbthambawita/divergent-net/releases/download/checkpoints_v1/best_checkpoint_TriUnet.pth
- https://github.com/vlbthambawita/divergent-net/releases/download/checkpoints_v1/best_checkpoint_unet_plusplus.pth
DivergentNet
Merging and predicting from divergent networks
Set following parameters in inference_from_divergentNets.sh
--input_dir <directory to input images>
--output_dir <directory to save predicted mask>
--chk_paths <path to pretrained checkpoints. You can provide single checkpoint path or multiple checkpoint paths. Use a space to seperate multiple checkpoint paths or '\' as the given example paths.>
Then run it:
bash inference_from_divergentNets.sh
For Windows users we provide an inference script that supports windows. You can run it without the bash. For example:
python inference_from_divergentNets.py --input_dir C:\Users\xxx\GitHub\divergent-nets\input --output_dir C:\Users\xxx\GitHub\divergent-nets\output --chk_paths C:\Users\xxx\OneDrive\Dokumente\GitHub\divergent-nets\checkpoints\best_checkpoint_Deeplabv3.pth
Sample predictions from different models used in DivergentNets and it's own output.
Citation
@inproceedings{divergentNets,
title={DivergentNets: Medical Image Segmentation by Network Ensemble},
author={Thambawita, Vajira and Hicks, Steven A. and Halvorsen, P{\aa}l and Riegler, Michael A.},
booktitle={Proceedings of the 3rd International Workshop and Challenge on Computer Vision in Endoscopy (EndoCV 2021)
co-located with with the 17th IEEE International Symposium on Biomedical Imaging (ISBI 2021)},
year={2021}
}
298 Dec 21, 2022
94 Dec 25, 2022
4 Jul 1, 2022
23 Mar 9, 2022
37 Oct 30, 2022
24 Sep 8, 2022
1 Jan 20, 2022
30 Nov 22, 2022
44 Jun 27, 2022
492 Jan 06, 2023
33 Dec 15, 2022
44 Dec 04, 2022
787 Dec 30, 2022
15 Dec 17, 2022
5 Nov 29, 2021
190 Dec 27, 2022
1.2k Dec 28, 2022
63 Aug 11, 2022
171 Jan 06, 2023
74 Dec 26, 2022
19 Nov 25, 2022
6 Oct 27, 2022
6 Jul 24, 2022
80 Dec 13, 2022
101 Nov 30, 2022
2 Jan 03, 2022
48 Dec 14, 2022
61 Jun 07, 2022
231 Jan 05, 2023