/OpenLabeling

Label images and video for Computer Vision applications

Primary LanguagePythonApache License 2.0Apache-2.0

OpenLabeling: open-source image and video labeler

GitHub stars

Image labeling in multiple annotation formats:

Citation

This project was developed for the following paper, please consider citing it:

@INPROCEEDINGS{8594067,
  author={J. {Cartucho} and R. {Ventura} and M. {Veloso}},
  booktitle={2018 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS)}, 
  title={Robust Object Recognition Through Symbiotic Deep Learning In Mobile Robots}, 
  year={2018},
  pages={2336-2341},
}

Latest Features

  • Jun 2019: Deep Learning Object Detection Model
  • May 2019: [ECCV2018] Distractor-aware Siamese Networks for Visual Object Tracking
  • Jan 2019: easy and quick bounding-boxe's resizing!
  • Jan 2019: video object tracking with OpenCV trackers!
  • TODO: Label photos via Google drive to allow "team online labeling". New Features Discussion

Table of contents

Quick start

To start using the YOLO Bounding Box Tool you need to download the latest release or clone the repo:

git clone --recurse-submodules git@github.com:Cartucho/OpenLabeling.git

Prerequisites

You need to install:

  • Python
  • OpenCV version >= 3.0
    1. python -mpip install -U pip
    2. python -mpip install -U opencv-python
    3. python -mpip install -U opencv-contrib-python
  • numpy, tqdm and lxml:
    1. python -mpip install -U numpy
    2. python -mpip install -U tqdm
    3. python -mpip install -U lxml

Alternatively, you can install everything at once by simply running:

python -mpip install -U pip
python -mpip install -U -r requirements.txt
  • PyTorch Visit the link for a configurator for your setup.

Run project

Step by step:

  1. Open the main/ directory

  2. Insert the input images and videos in the folder input/

  3. Insert the classes in the file class_list.txt (one class name per line)

  4. Run the code:

  5. You can find the annotations in the folder output/

    python main.py [-h] [-i] [-o] [-t] [--tracker TRACKER_TYPE] [-n N_FRAMES]
    
    optional arguments:
     -h, --help                Show this help message and exit
     -i, --input               Path to images and videos input folder | Default: input/
     -o, --output              Path to output folder (if using the PASCAL VOC format it's important to set this path correctly) | Default: output/
     -t, --thickness           Bounding box and cross line thickness (int) | Default: -t 1
     --tracker tracker_type    tracker_type being used: ['CSRT', 'KCF','MOSSE', 'MIL', 'BOOSTING', 'MEDIANFLOW', 'TLD', 'GOTURN', 'DASIAMRPN']
     -n N_FRAMES               number of frames to track object for
    

To use DASIAMRPN Tracker:

  1. Install the DaSiamRPN submodule and download the model (VOT) from google drive
  2. copy it into 'DaSiamRPN/code/'
  3. set default tracker in main.py or run it with --tracker DASIAMRPN

How to use the deep learning feature

Download the pre-trained model by clicking this link http://download.tensorflow.org/models/object_detection/ssdlite_mobilenet_v2_coco_2018_05_09.tar.gz and put it into object_detection/models. Create the models folder if necessary. Make sure to extract the model.

Note: Default model used in main_auto.py is ssdlite_mobilenet_v2_coco_2018_05_09. We can set graph_model_path in file main_auto.py to change the pretrain model

  • Using main_auto.py to automatically label data first

    TODO: explain how the user can

GUI usage

Keyboard, press:

Key Description
a/d previous/next image
s/w previous/next class
e edges
h help
q quit

Video:

Key Description
p predict the next frames' labels

Mouse:

  • Use two separate left clicks to do each bounding box
  • Right-click -> quick delete!
  • Use the middle mouse to zoom in and out
  • Use double click to select a bounding box

Authors

  • João Cartucho

    Feel free to contribute

    GitHub contributors