Semantic Segmentaion for Weizmann Horse Database

[Homework] Fanpeng Meng's Final work for Visual Cognition Engineering 2022-Spring, School of AIA, HUST.

Summary

In this work, I implied semantic segmentation model UNet, Unet++ and EMANet on Weizmann Horse Dataset, acheiving the target MIOU and BIOU in part of the experiments.

UNet [paper] UNet++ [paper] [code] EMANet [paper] [code]

[DATASET] Weizmann Horse [Kaggle]

[EVALUATION] BIOU [paper]

Unet & Unet++

EMANet

Installation

Using commands to install the environment:

pip -r requirements.txt

Data Preparation

Prepare your dataset as below:

dataset 
  └──── weizmann_horse_db
    ├──── horse 
    │ ├──── horse001.png 
    │ ├──── horse002.png 
    │ ├──── horse003.png
    │ ├──── ... 
    │ └──── horse327.png 
    └──── mask 
      ├──── horse001.png 
      ├──── horse002.png 
      ├──── horse003.png
      ├──── ... 
      └──── horse327.png

Training and Testing

Training and testing on UNet, configuring the training in config/unet_config.py:

python main_unet.py

Training and testing on UNet++, configuring the training in config/unet_config.py:

python main_unetpp.py

Training and testing on EMANet, configuring the training in config/emanet_config.py:

python main_emanet.py

I trained the model on a single NVIDIA A30 (24 GB). 250 epochs training takes about 10 minutes for UNet, 20 minutes UNet++ and EMANet. The current bottleneck is the parallel computing, I didn't apply parallel training for the reason that the training scale is quite elite, it's not worthy to do that.

Result and Explaination

The result is shown below:

MIOU:

Model	1 Epoch	20 Epoch	Best
UNet	0.576	0.904	0.914
UNet++	0.588	0.817	0.889
EMANet	0.617	0.791	0.806

BIOU:

Model	1 Epoch	20 Epoch	Best
UNet	0.310	0.717	0.745
UNet++	0.164	0.536	0.701
EMANet	0.198	0.459	0.493

Best means the best performance in 250 epochs
The results are shown in detail in my report (CN). I didn't achieve perfect performance on EMANet, and the analyze was also mentioned in the report.
My weight of models was saved in my Google Drive. I saved the weight every 50 epochs, as well as the weight of best model.
My tensorboard events of the training was also provided in my Google Drive. You can review my training.
Support Material: Google Drive

Contract

Mail me at HUST-Mail or Gmail

Reference

The following works are appreciated for great credit to my work:

@inproceedings{ronneberger2015u,
  title={U-net: Convolutional networks for biomedical image segmentation},
  author={Ronneberger, Olaf and Fischer, Philipp and Brox, Thomas},
  booktitle={International Conference on Medical image computing and computer-assisted intervention},
  pages={234--241},
  year={2015},
  organization={Springer}
}

@incollection{zhou2018unet++,
  title={Unet++: A nested u-net architecture for medical image segmentation},
  author={Zhou, Zongwei and Rahman Siddiquee, Md Mahfuzur and Tajbakhsh, Nima and Liang, Jianming},
  booktitle={Deep learning in medical image analysis and multimodal learning for clinical decision support},
  pages={3--11},
  year={2018},
  publisher={Springer}
}

@inproceedings{li2019expectation,
  title={Expectation-maximization attention networks for semantic segmentation},
  author={Li, Xia and Zhong, Zhisheng and Wu, Jianlong and Yang, Yibo and Lin, Zhouchen and Liu, Hong},
  booktitle={Proceedings of the IEEE/CVF International Conference on Computer Vision},
  pages={9167--9176},
  year={2019}
}

mfp0610 / semantic-segmentaion