- Intel I7-7655K, 32G DDR4, Nvidia Titan X
- Xubuntu 16.04.1, CUDA 8.0, cudnn 5.1
1. Install the following components
- python 2.7
- numpy
- scipy
- pandas
- matplotlib
- scikit-learn
- pillow
- tensorflow
- keras
- h5py
- pydot
├── data
│ ├── raw
│ ├── test
│ └── test
├── features
├── logs
├── models
├── submitions
├── utils
├── weights
├── submit.py
└── train.py
2. Download model weight files to 'weights/imagenet'
3. Download kaggle compitition data
4. Train single models and get results
- Run 'python train.py -m [inception|resnet50|xception]' to get transform learning model
- Run 'python finetune.py -m [inception|resnet50|xception]' to fine tune model
- All model weight files are in the folder 'weights'
- Run 'python submit.py -m [inception|resnet50|xception]' to get submition
- All submitions are in the folder 'submitions'
5. Train blend model and get results
- Run 'python blend.py -m xception resnet50 inception' to get blend result with validation
- Run 'python blend.py -m xception resnet50 inception -a true' to get blend result without validation