Keras Object Detection API with YOLK project

YOLK: Keras Object Detection API

You Only Look Keras

Every year newly developed Object Detection architectures are introduced, but even applying the simplest ones has been something with, or perhaps more than, a big hassle so far. YOLKYou Look Only Keras is an one-stop Object Detection API for Keras, which is built as a part of 2019 Open Source Contributhon. It shares the idea of what Keras is created for: Being able to go from idea to result with the least possible delay is key to doing good research. With a few lines of codes, you can set up and apply one of the best performing models to your own datasets. We hope this can help everyone train their own Object Detection models painlessly.

Detailed instructions will be updated soon.🤓

Contents

• Directory Structure
• Installation
• Quick Start
• Dependencies
• Release information
• Contributors

Directory Structure


KerasObjectDetector
├── datasets
├── docker
├── examples
│   ├── ResNet50RetinaNet
│   ├── YOLOv3_Prediction
│   ├── YOLOv3_Training
│   ├── retinanet_inference_example
│   ├── retinanet_training_example
│   ├── ssd_inference_example
│   └── ssd_training_example
├── keras_retinanet
├── keras_ssd
├── keras_yolov3
├── res
├── tests
├── yolk
├── .gitignore
├── LICENSE
├── README.md
├── setup.cfg
├── setup.py

Installation (On Linux)


Download YOLK API. This API helps setting up development environment for Object Detection tasks. Run the following commands in terminal.

  # Download YOLK API
  $ git clone https://github.com/KerasKorea/KerasObjectDetector.git
  $ cd KerasObjectDetector

  # If there is no 'setuptools' in docker, please download this package.
  # pip install setuptools
  # install library
  $ apt-get install libatlas-base-dev libxml2-dev libxslt-dev python-tk
  
  # build setup codes
  # ./KerasObjectDetector
  $ python setup.py install

If you want to run the API on a docker, download our pre-built docker image from Docker HUB and run.

  #  pull yolk docker image
  $ docker pull kerasyolk/yolk

  # run yolk
  $ docker run --name=yolk -Pit -p 8888:8888 -p 8022:22 kerasyolk/yolk:latest

  # running jupyter-notebook
  $ jupyter-notebook

Quick Start


You can test your own image with a Quick Start :)

Dependencies


Release information


ver 1.0.0 (November 20, 2019)

The initial version of YOLK API has been completed. This release includes:

1. Three Object Detection models and one data generator. (More models will be included in a future version). The generator converts datasets into a suitable format for a selected model.
   • Object Detection models : SSD, YOLOv3, RetinaNet
   • Dataset and data generator : PASCAL VOC2012, COCO, Custom dataset
     ㄴ Yolk’s dataset downloader is 3X faster than existing downloaders.
2. Docker files for quick development environment setup.
3. Easy and detailed Object Detection tutorials (SSD+VOC2012).

Contributors


Thanks goes to these beautiful people (github ID) : @fuzzythecat, @mijeongjeon, @tykimos, @SooDevv, @karl6885, @EthanJYK, @minus31, @young917, @mike2ox, @hngskj, @hics33, @aaajeong, @parkjh688, @Uwonsang, @simba328, @visionNoob, @Naruu, @melonicedlatte, @magh0, @ahracho