Google on Monday released the latest version (version 4) of ‘Open Images‘, enabling Data Scientists and AI professionals around to world to feed their systems with data, generating models that learn – and opened up a major new visual object competition.
What is Open Images?
Open Images, is a gargantuan image database initially released by Google in 2016. It can be used for building out Data Science and Artificial Intelligence algorithms within a range of contexts. A key use case of this database is consequently Machine Learning, which essentially feeds large sets of data to a system which will derive meaningful insights to make informed solutions to problems.
This is in contrast to traditional computer software, which takes gives a program data to calculate outputs from without true learnings.
Illustratively, this is the difference between asking a computer to calculate ‘1 + 1’ as opposed to ‘What is the best equation to use for a given problem’. The tech giant said: “We are happy to announce Open Images V4, containing 15.4 million bounding-boxes for 600 categories on 1.9 million images, making it the largest existing dataset with object location annotations”.
Fundamentally, this is very useful for the AI community, who will be able to experiment with the data, and it will be interesting to see the real-world applications that emerge from it.
Google’s Investment in AI and ML
With notable acquisitions of cutting-edge AI firms such as DeepMind, as well as lesser known acquisitions such as space-based geocomputation firm Terra Bella, Google’s investment in AI and Machine Learning is evident.
This could explain why it is releasing such vast amounts of data for shared usage, which is not only common in the software industry (known as open sourcing), but also beneficial for them through the research insights that can be gleaned.
In the light of this, the firm also noted as part of the release “We hope that the very large training set will stimulate research into more sophisticated detection models that will exceed current state-of-the-art performance”. This acknowledgement of the benefits the firm will derive from releasing this data shows the great utility of open sourcing.
Google sets challenge for AI community
The release of Open Images coincides with a challenge set by Google to be held at a major Computer Vision conference in Munich, Germany this year, that illustrates how such collaboration and community-based problem solving is key in Google keeping ahead of its competition.
This challenge is unique in several ways and includes:
- 12.2 million bounding-box annotations for 500 categories on 1.7 million training images,
- A broader range of categories than previous detection challenges, including new objects such as “fedora” and “snowman”.
- In addition to the object detection main track, the challenge includes a Visual Relationship Detection track, on detecting pairs of objects in particular relations, e.g. “woman playing guitar”.
The training set is available now. A test set of 100,000 images will be released on July 1st 2018 by data science platform Kaggle, Google said. Deadline for submission of results is on September 1st 2018.
The company’s Vittorio Ferrari, a machine perception research scientist, said in a blog: “We hope that the very large training set will stimulate research into more sophisticated detection models that will exceed current state-of-the-art performance, and that the 500 categories will enable a more precise assessment of where different detectors perform best. Furthermore, having a large set of images with many objects annotated enables to explore Visual Relationship Detection, which is a hot emerging topic with a growing sub-community.”
The competition comes as Computer Business Review revealed that Microsoft was also making progress in automated image segmentation, through trials held in collaboration with the retail industry.