Flare removal models for LG dacon competition.
${POSE_ROOT}
|-- data
|-- train
| |-- input
| | |-- inp1.png
| | |-- inp2.png
| | |-- ...
| |-- target
| | |-- tar1.png
| | |-- tar2.png
| | |-- ...
|-- test
| |-- input
| | |-- inp1.png
| | |-- inp2.png
| | |-- ...
| |-- target #option
| | |-- tar1.png
| | |-- tar2.png
| | |-- ...
This repository use 2 stream models. You have to train each model seperately.
python run/train.py
python run/train_merge.py
python run/test_merge.py
Input | Prediction | Target | 1.0 scale attention | 0.5 scale attention | 0.25 scale attention | 0.125 scale attention