Project description
A closer look at a new image database platform
The emergence of improved machine learning (ML) algorithms, progress in computing power and greater accessibility through big data have boosted artificial intelligence (AI). Also, its use by researchers and application developers for computer vision is also on the rise. Despite several image databases to train ML algorithms to recognise a series of generic categories, there is a lack of image databases comprising more specific features of interest. What is more, there is no platform yet that offers a streamlined, standardised method to install and run an image classification campaign. The EU-funded PPP project will develop an innovative and commercially self-sustained platform based on the crowdsourcing game Picture Pile.
Objective
There has been tremendous progress in artificial intelligence (AI) in many different fields due to improved machine learning algorithms, advances in computing power and the availability of big data. With machine learning platforms like TensorFlow (Google), PyTorch (Facebook), Azure (Microsoft) or CoreML (Apple), researchers and application developers can now use AI for applications such as computer vision. While there are many image databases available such as Imagenet, which can be used to train machine learning algorithms to recognize a set of generic categories, e.g. cats, there is still a lack of image databases containing more specific features of interest, e.g. crop types. Moreover, there is currently no platform that offers a streamlined, standardized approach to setting up and running an image classification campaign to ingest the millions of photographs currently collected by many people around the world. The value proposition of the ERC PoC Picture Pile Platform (PPP) is to develop an innovative, commercially self-sustaining platform that uses the crowdsourcing game Picture Pile to rapidly classify images for machine learning purposes. After a pile of images has been sorted, the image classifications will be made publicly available. A number of premium services will be added to the Picture Pile Platform, which will make the platform self-sustaining. Campaign creators will be able to pay the crowd (with a small share paid to the Picture Pile Platform) for sorting the images. Paid advertisements will be possible through our social media channels as well as paid for space on the main Picture Pile page to attract more users. For a small fee, users will be able to use the Picture Pile Cloud for specific computer vision tasks using the machine learning models that will be built and trained with the Picture Pile data sets. The Picture Pile Platform has the potential to become a self-sustaining hub for crowdsourcing image classifications for machine learning.
Fields of science
- natural sciencescomputer and information sciencesdatabases
- natural sciencescomputer and information sciencesdata sciencebig data
- natural sciencescomputer and information sciencesartificial intelligencecomputer vision
- agricultural sciencesagriculture, forestry, and fisheriesagriculture
- natural sciencescomputer and information sciencesartificial intelligencemachine learning
Programme(s)
Funding Scheme
ERC-POC - Proof of Concept GrantHost institution
2361 Laxenburg
Austria