competetion named "Steel Defect Detection" on kaggle please download dataset from the above
if you want to check total number of training data and visualize some of them
python3 ./datavis.py
you can choose following models and loss functions, please type following keywords in ()
models : -m (default = unet)
- UNet (unet)
- ResNet-based Unet (resnetx: x is num layers of ResNet)
- ResUNet-a (resunet-a)
-m <MODEL NAME>
loss functions : -l (default = BCE)
- BCE (BCE)
- Dice (Dice)
- DiceBCE (DiceBCE)
- IoU (IoU)
- Focal (Focal)
- Tanimoto (Tanimoto)
-l <FUNCTION NAME>
Then, you can train model like this
$ cd src
$ python3 ./main.py -m <MODEL NAME> -l <FUNCTION NAME>