Setup

Before installation create and activate virtual environment

python3 -m venv venv
source venv/bin/activate

Install the dependencies

pip install -r requirements.txt

Training

For training run jupyter notebook

Additional instructions

Data preparation

We use the NUS-WIDE dataset for this tutorial. Instead of parsing Flickr for image downloading we use a dump from this github repository Download and extract it.

Also, we added pre-processed annotations:
nus_wide/train.json
nus_wide/test.json

If you want to create them yourself, run the command:

python split_data_nus.py -i images

where -i images is the path to the folder with extracted images

Subset creation

You can train the model for the entire data set, but it takes a lot of time. For this tutorial we use part of this data.

For subset creation run the command:

python create_subset.py -i images

where -i images is the path to the folder with extracted images

Additional options:

python create_subset.py -h
usage: Subset creation [-h] -i IMG_PATH [-v VAL_SIZE] [-t TRAIN_SIZE]
                       [--shuffle] [-l LABELS [LABELS ...]]

optional arguments:
  -h, --help            show this help message and exit
  -i IMG_PATH, --img-path IMG_PATH
                        Path to the "images" folder
  -v VAL_SIZE, --val-size VAL_SIZE
                        Size of the validation data
  -t TRAIN_SIZE, --train-size TRAIN_SIZE
                        Size of the train data
  --shuffle             Shuffle samples before splitting
  -l LABELS [LABELS ...], --labels LABELS [LABELS ...]
                        Subset labels