Recognizing that image datasets are geographically skewed towards North America and Europe, Google is concerned to find machine learning methods that are more inclusive. The Inclusive Images Contest has just launched on Kaggle.

Announcing the contest on the Google AI blog, Tulsee Doshi reminds us that the availability of large image datasets such as ImageNet, has been an important driving factor in the progress made recently in computer vision. However the lack of geo-diversity in these datasets means that models based on them don't perform well on images drawn from other geographical regions. This example, which shows a standard open-source image classifier trained on the Open Images dataset and fails to properly apply “wedding” related labels to images of wedding traditions from different parts of the world, is used to make this point:

Explaining why Google has partnered with the Conference on Neural Information Processing Systems (NIPS) Competition Track in the Inclusive Images Challenge, Doshi writes:

While Google is focusing on building even more representative datasets, we also want to encourage additional research in the field around ways that machine learning methods can be more robust and inclusive when learning from imperfect data sources. This is an important research challenge, and one that pushes the boundaries of ways that machine learning models are currently created. Good solutions will help ensure that even when some data sources aren’t fully inclusive, the models developed with them can be.

The contest, which uses the  Open Images dataset, started on Kaggle September 5th with the available training data and first stage Challenge data set. The deadline for submitting your results will be Monday, November 5th. The second stage Test data set will be released on Tuesday, November 6th and its deadline in November 12th.

The results of the competition will be presented at the 2018 Conference on Neural Information Processing Systems, being held in Montreal, Canada in December and the five top-ranking competitors will be awarded travel grants of $5,000 to attend the conference and present their Inclusive Images solution, which have to be open sourced, as part of the NIPS 2018 competition track. 

Teams can have up to eight members and the team merger deadline is the same as the entry deadline - October 29th. For more details and timelines, visit the Inclusive Images Competition website. 

More Information

Introducing the Inclusive Images Competition

Inclusive Images Challenge

Conference on Neural Information Processing Systems

NIPS 2018 Competition Track

Related Articles

Google Discovers That Neural Networks Just Need More Data 

Kaggle Enveloped By Google Cloud

Higgs Boson Machine Learning Challenge On Kaggle

Fun Kaggle Challenge To Tell Dogs From Cats 

ImageNet Training Record - 24 Minutes 

Google's New Contributions to Landmark Recognition 

Malmo AI Available On Github 

To be informed about new articles on I Programmer, sign up for our weekly newsletter, subscribe to the RSS feed and follow us on Twitter, Facebook or Linkedin.

Comments

Make a Comment or View Existing Comments Using Disqus

or email your comment to: comments@i-programmer.info