What data makes sense for the DataUnion.app Image & Annotation Vault?
During the challenges all data that complies with the rules is fine to be uploaded but to earn royalties from your data, the data also has to be bought on the Data Portal. We will make it visible there how often each data is being sold to indicate which data makes sense to be created. But of course we already have insights into which data is promising - object data is the most promising. So if there is one or more objects clearly visible in the image then this is valuable. This also helps to annotate this information well. Another factor is objects in unexpected forms and in interesting situations e.g. a traffic sign with a graffiti on it, roadkill, kissing cats or a duck sitting on a car. But this will be a dynamic that plays out via the Data Portal for now there is not much in the dataset, get creative and figure out your nieche to fill.
Another way to get rewards after the challenges are data bounties. To take part in a data bounty you have to add the correct tag to the upload or verification. To test how successful they are we are adding a number of data bounties during the challenges. This gives us insight into how well these bounties can be fullfilled and also will fuel algorithms that we want to develop/improve. Additionally we chose data bounty categories that can highlight the diversity and global present of our community.
To filter out personal information from image data we need to train an algorithm that is based on this personal information. On the page about personal information this data and the different catogries of personal information are defined. This is the first bounty that we are looking for. Use the checkboxes on the bottom of the upload and verification pages to add the correct tags for this bounty.
This algorithm will be used by ourselves in the future and will be continuously updated as we want to make it 100% sure to not show personal information in the Data Portal. A partial algorithm of this will be the anonymization of faces, this will be the first initial algorithm offering that we will launch on the Data Portal. Contributors to this data will then also get rewarded with datatokens of this algorithm. We are sure that our DataUnion.app community from around the world will be able to create a diverse dataset that will yield a highly succesful algorithm.
Please note that we will not offer this information on the Data Portal without taking precautions as the correct handling of personal information is of highest importance to us. The tag for this bounty is: anonymization bounty
Traffic signs bounty
In collaboration with Evotegra we are working on Swipe-AI, a new way of annotating images, by comparing a target class image with a crop of the real image. If they match the contributor swipes right, otherwise left. So we are looking for more images around the world to make have data that can help this algorithm to generalise. The tag for this bounty is: traffic sign bounty
Food is different around the whole world and this is another excellent usecase to showcase the diversity of our community. Take a picture of your meals, fruits, drinks, and whatever else you deem as food and upload the pictures. The tag for this bounty is: food bounty
Cigarette butts on the beach bounty
This is the most specific bounty that we are investigating. Here we are looking for images of cigarette butts on the beach, preferable from directly above them in 20-40cm or 8-16in distance. And only if they can be reached by a robot actuator. This data is used to train the robot of Project.BB to detect and collect cigarette butts on the beach. The tag for this bounty is: project.bb bounty
And here is a picture of the robot to give you a better impression: