Ternausnet with tensorflow implementaion (and it contains few lines of pytorch code). Original TernausNet was implemented with pytorch.
This Project was developed for TGS Slat identification Challenge. So, If you want this model used for other domain or data, you will change some code. (like input and output shape).
- Python 3.5 or higher.
- CUDA 9.0 compatible GPU
install package dependencies like below.
> pip3 install -r requirements.txt
- tensorboard==1.9.0
- tensorflow-gpu==1.9.0
- tqdm==4.24.0
- numpy==1.15.0
- pandas==0.23.4
- sklearn
- batchup==0.2.0
- torchvision==0.2.1
- https://www.kaggle.com/c/tgs-salt-identification-challenge/data
- Download dataset from above link. and then, extract every zip files.
- Make directory named ‘input’. and move all dataset into 'input' directory.
.
├── main.py ( code that has main function )
├── metrics.py ( some metrics )
├── model.py ( model codes )
├── requirements.txt (Package dependencies )
├── utils.py ( utility codes )
├── logs (log directory generated by model )
└── train ( tensorboard train log)
└── validation (tensorboard validation log)
├── model (checkpoint directory generated by model )
> python3 main.py --mode train --epochs [EPOCHS] --batch_size [BATCHSIZE] --checkpoint [CHECKPOINT]
> python3 main.py --mode test --batch_size [BATCHSIZE] --checkpoint [CHECKPOINT]
The Project is released under MIT See LICENSE for details.