cna74 / Basilia

object detection with tensorflow-keras based on Open-Image

Basilia

An API for Open-Image-V4 to gather your desired images with bboxes in just TWO LINES OF CODE!

Downloading DataSet

if you want to use online just download csv and json

• csv:

  train-annotations-bbox.csv	train-images-boxable.csv
  validation-annotations-bbox.csv	validation-images.csv
  test-annotations-bbox.csv	test-images.csv
  class-descriptions-boxable.csv	json_hierarchy

• zip *: download zip file if you want to use basilia in offline mode

  train_00	train_01	train_02
  train_03	train_04	train_05
  train_06	train_07	train_08
  validation	test

  download and extract them

• store files like this, stared items are't necessary for online use

└── Open_Image
    ├── class-descriptions-boxable.csv
    ├── bbox_labels_600_hierarchy.json
    ├── Test
    │   ├── test-annotations-bbox.csv
    │   ├── test-images.csv
    │   └── test       *
    ├── Train
    │   ├── train_00   *
    │   ├── train_01   *
    │   ├── train_02   *
    │   ├── train_03   *
    │   ├── train_04   *
    │   ├── train_05   *
    │   ├── train_06   *
    │   ├── train_07   *
    │   ├── train_08   *
    │   ├── train-annotations-bbox.csv
    │   └── train-images-boxable.csv
    └── Validation
        ├── validation *
        ├── validation-annotations-bbox.csv
        └── validation-images.csv

create environment [optional]

• conda create -n ElBasil python=3.6
• source activate ElBasil

clone Basilia

• clone repo git clone https://github.com/cna74/Basilia.git
• install requirements pip install -r Basilia/requirements.txt

install object-detection-api

• clone repo git clone https://github.com/tensorflow/models.git
• move to research directory cd models/research/
• install object-detection-api python3 setup.py install

edit config.py file

• open Basilia/utils/config.py

  DATA_DIR="<path-to-open-image-directory>"
  # if you downloaded zip files and extracted them
  RESOURCE="jpg"
  # if you just downloaded csv files
  RESOURCE="csv"

• from directory Basilia/:

  from Basil import Finder
  # find Punching bags
  Finder(subject="Punching bag", out_dir="/home/cna/Desktop/", automate=True)

  

  from Basil import Finder
  # find fox, tiger and jaguar
  Finder(subject=("fox", "tiger", "Jaguar"), out_dir="/home/cna/Desktop/", automate=True)

  

  from Basil import Finder
  # just count all type of fruits, won't extract them
  Finder(subject="fruit", out_dir="/home/cna/Desktop/", just_count=True, automate=True)
  '''  just count images:
            Images  Objects
  Train        13271        0
  Validation     786        0
  Test          2545        0
  '''

more examples in https://github.com/cna74/Basilia/blob/master/Examples.ipynb

Tutorial

part 1

part 2