README.md

Catalyst classification example with training and infer stage

Installation:

To install all requirements run pip install -r requirements/requirements.txt

Training

1. Unpack dataset into "data" folder. In my example I use 2 classes. You should have next structure.

data
├── test
    ├── 0001.jpg
    ├── 0002.jpg 
    └── ... 
└── train
    ├── 0_class
        ├── 0001.jpg
        ├── 0002.jpg 
        └── ... 

    ├── 1_class
        ├── 0001.jpg
        ├── 0002.jpg 
        └── ... 
    ...

    ├── n_class
        ├── 0001.jpg
        ├── 0002.jpg 
        └── ... 

2. Rename folders in train to 0,1,... and so on.

data
├── test
    ├── 0001.jpg
    ├── 0002.jpg 
    └── ... 
└── train
    ├── 0
        ├── 0001.jpg
        ├── 0002.jpg 
        └── ... 

    ├── 1
        ├── 0001.jpg
        ├── 0002.jpg 
        └── ... 
    ...

    ├── n_class
        ├── 0001.jpg
        ├── 0002.jpg 
        └── ... 

3. Run sh bin/prepare_data.sh to prepare data. It splits you train folder into train and val and preprocess test for infer stage.
4. Run catalyst-dl run -C configs/_common.yml configs/main.yml --logdir=baseline If you want use more than 2 classes change &num_classes 2 to custom number in infer.yml and main.yml files. (Important! Not set to 1 class, this feature isn't supported)
5. Run catalyst-dl run -C configs/_common.yml configs/infer.yml --logdir=baseline --logdir=baseline --autoresume=last to make predictions. It shows prediction for each file and dump it into 'infer_pred.txt' file in 'baseline' folder