Welcome to Kootstrap

Kootstrap is a bootstrap to Keras. It is a technique of compile and loading a datasets into a Keras application by means of a few initial instructions that enable the introduction of the rest of the program from an input device

Get Code

If you want to get code or contribute, go to Kootstrap GitHub.

Commands

Bellow you find how create a dataset, a subset, crawls images and execute a trainig your own model.

Create a dataset

Create a dataset with two classes.

cd maker/
python Main.py --mode dataset --dataset_name graffiti --classes graffiti,street

Create a subset

Create a subset of this dataset with 90% original images.

python Main.py --mode compiler  --dataset_name graffiti --subset_name graffiti_per90_porp_default --per_images 90

Crawls images

Crawls images to each class from Flickr. This seed the dataset and compile the subsets.

cd ../crawler/
python Main.py --mode crawler,dataset --dataset dataset_example --classes graffiti,street --flickr_tags graffiti,street --num_images 100

Execute a training

Execute a training with finetuning in Imagenet Model.

cd ../trainer/
python Main.py --model_name model_example_1 --load_data dataset_example

Test the predictions

Test the predictions on model with set of test compileted in graffiti_per90_porp_default.

cd ../tester/
python Main.py --model_name model_example_1 --load_data graffiti_per90_porp_default

Compile a 1-Top charts

Compile a 1-Top with a histogram from test results.

cd ../analyzer/
python Main.py --model_name top --test_name testing_imagenet_test_set

Tools

Migrate datasets

if you have a dataset and want migrate try:

cd ../tools/
python Main.py --mode migrate --path_origin <PATH_FOLDER_WITH_CLASSES> --path_destiny <PATH_TO_KOOTSTRAP_FOLDER>

if you want create a subset or recovery the metadata.json try:

cd ../tools/
python Main.py --mode fix --path_origin <PATH_TO_SUBSET_OR_DATASET>

Project layout

data/           # folder with all data generate by applications.
    configs/
    models/
    datasets/
    tests/
    others/
applications/   # applications suches crawler, analyzers etc
    analyzer/
    crawler/
    maker/
    system/
    tester/
    tools/
    trainer/
docs/           # documentation of kootstrap
mkdocs/         # generator of documentation

Graphium

Try use models trained on Kootstrap on your Graphium application.