The story and Future of ImageAI – One year Anniversary

Source: Deep Learning on Medium

It is with utmost excitement that I write this piece, not only because the ImageAI project has made incredible impact since it’s launch on the 22nd of March 2018, but because of the innovations that will be made possible by the project in years to come. It’s been been a year and 2 months since We announced the ImageAI, a computer vision library we built to enable developers of any level of expertise to access and integrate state-of-the-art computer vision functionalities powered by deep learning. Today, ImageAI is being used by over 65,000 developers, engineers, college students, research institutes, businesses and emerging startups around the world to build incredible ideas and innovations.

In this post, I will be addressing 3 points on the project which are the

– Global coverage in numbers.

Prominent applications and case studies

Timeline and features for the Next Version

(1) Global Coverage

When we published the ImageAI library coupled with it’s documentation in English language and various tutorials highlighting the implementations of its functionalities, we had no idea the extent at which the project whose source code originated from our computer systems will be used across various countries and various industries around the world. Since the past one year and two months, ImageAI

  • has been installed over 65,000 times ( link )
  • used in 60+ countries around the world.
  • been featured on 200+ publications and research papers globally
  • official tutorials have been read over 420,000 times
  • have been featured in scores of video tutorials with tens of thousands of views (Google Videos) (YouTube)

(2) Case Studies

Over the past 1 year, I have received over 1000 emails from various developers, engineers, students and researchers based in various parts of the world on how they have been applying ImageAI to power incredible innovations and and projects they are working on. Interestingly, most of these individuals and groups are newbies in Python programming and some of them have no prior knowledge of computer programming when they started using the ImageAI library. With the simple, well documented but yet powerful recognition, detection and analysis features ImageAI provides, the 5 case studies listed below were made possible. Soon I will be publishing comprehensive articles detailing each case study.

  • Professor Sandy from the University College Dublin, Republic of Ireland.

Professor Sandy Wilkinson is an historian working on preserving millions of historical ornamentation and illustrations printed in Europe before the year 1800. His department had digital copies of these illustrations but had been having problems manually sorting them due to the enormous manual task involved. Despite being a newbie in Python programming, Professor Sandy was able to train a custom recognition model in just 5 lines of code using ImageAI which automated the sorting of millions of the digital copies, saving him thousands of hours of manual work. In his words, he said

As an historian, I am very new to machine learning and indeed programming, but found your github software fantastically helpful! I have trained a test RESNET50 model following your detailed and very helpful instructions. It produces astonishingly accurate results.

  • Asst. Professor Dorte from the University of Wisconsin-Madison, USA

Assistant Professor Dorter Dopfer is an Epidemiologist in the Department of Medical Science working extensively on infectious diseases and population health of animals. She has been using ImageAI project to easily train custom recognition models to enhance her research work. In her words , she said

I found your documentation about ImageAI and wanted to express my respect and compliments for the work that you have done! Very helpful for me who is training in applications of Computer Vision”.

  • Feinstein Institutes for Medical Research, USA.

This case study is one those I am most proud and passionate about. The Feinstein Institute of Medical Research from Northwell Health is an institution of over 4000 brilliant and dedicated scientists in the the US that have made incredible breakthroughs and published groundbreaking research in the treatment of lupus, arthritis, sepsis, cancer, psychiatric illness and Alzheimer’s disease. One of their notable breakthroughs is with Ian Burkhart, a quadriplegic with spinal injury who now has the opportunity to move his limbs using the brain-computer interface developed by the institute.

Interestingly, the Center Head for Bioelectronic Medicine at the institute, Chad Bouton reached out to our team where he mentioned they have been actively using ImageAI’s video detection and analysis feature as part of their research work. In his words, he said

First, let me say I’m very impressed with your work and I’m particularly impressed that you are self-taught in several difficult to learn areas. Kudos to you for taking the initiative and your passion for what you do. Just tried your ImageAI and it’s great!”.

  • 14-year Old Omshreshti from India

My joy knew no bounds on the weekend of March 2, 2019 when I heard the story of 14-year old Omshreshti, a brilliant and one of the most talented teenagers in India. Omshreshti participated in an Raspberry Pi Anniversary IoT Science Fair in India and until the final moments of the competition, he has not been able to figure out how to deploy state-of-the-art computer vision into his project. A brilliant data science by the name Soma Bhadra then suggested that Omshreshti implements ImageAI detection functionalities into his projects.

Within minutes, the young teen Omshreshti had his project running one of the most advanced object detection algorithms on his project . The most beautiful part was that his project became the most popular at the IoT Science Fair and his friends hoisted him on their shoulders & cheered him apparently! In his words, Omshreshti said

Thank You so much! Your ImageAI library helped me submit & be the most popular too. Hope more people benefit from it & contribute to it too. I hope I learn enough to contribute also.

  • Andrew Scott Meyers and his team, Masters student at University of California, Berkeley, USA

With the limitless possibilities that deep learning and computer vision have, Andrew Scott Meyers, a Masters Student at UC Berkeley, USA used ImageAI to build a waste scanner project which they showcased at the the just concluded UC Berkeley Design Innovation Showcase. With limited knowledge of computer programming and machine learning, this passionate team of researchers were able to leverage state-of-the-art recognition and detection algorithms to identify waste and manage waste disposal in solving global environmental issues; all possible using the easy to use functionalities of ImageAI. In his words, Andrew said

There was plenty of interest in this project at the showcase and most were impressed with the ease of implementing our idea using the ImageAI library especially considering our goal of a low-fidelity prototype and our minimal experience with python, computer vision, and machine learning. We are immensely thankful for ImageAI, and are excited to point others to your work.

(3) Future Versions of ImageAI

We have been inspired by the stories, use cases, videos, tutorials, contents and research work that has been generated with ImageAI . It has been 10 months since the last update was made to ImageAI (v2.0.2) . We have reviewed all the comments, suggestions and feedback from the community of developers using and supporting this project. We will also like to use this opportunity to thank them for their patience in the past months waiting for the next version. Our team had been working on various projects including ImageAI. The next version of ImageAI that will be released latest by June 15, 2019 will provide the following functionalities:

  • Setting video detection duration limits
  • Save custom trained models in full, ensuring custom recognition without specifying model type
  • Export custom trained model to Tensoflow .pb format
  • Continuous model training from previously saved model files
  • Transfer learning for training with few number of images
  • Run multiple custom recognition models
  • Export model to DeepStack API format
  • Custom object detection training and inference using YOLOv3

In preparation for this release, we have launched the ImageAI forum to ensure we an connect with the community of developers using ImageAI, discuss on bugs, feature request, classes and functions clarifications, tutorials, documentations, sample applications and use case stories.

As always, the new release will be fully documented, sample codes for all new features will be provided and extensive number of tutorials will be be published. If you will like to reach out to us and discuss on the project, feel free to join the ImageAI forum linked above. If you desire to contact us directly on your team or individual project, you can always reach to me via my email and social profiles below.