Swipe-AI - Mass Scale Data Farming

###Key Project Data

0x61B15998893cC746B46C08FEdEE13a0d1b33bBa9 (Evotegra)

ty (DataUnion.app)
The grant will be shared 50:50

Project Overview

The proposal in one sentence:

Enable potentially every smartphone user to get rewards based on datatokens by curating AI pre-labelled data.

  • Which category best describes your project? Pick one or more.

  • [x] Build / improve applications or integrations to Ocean

Description of the project:

Labelling data on a smartphone is not just inefficient. It is also introducing a human perception bias into the data, preventing AI from leveraging its full potential. And while datasets are getting continuously more complex, any additional complexity will deteriorate the quality achievable by humans. To solve this issue we introduce a revolutionary approach allowing humans to scale 100x while significantly improving the quality and hence the value of the data.

We present Swipe-AI. The first human and AI collaboration at mass scale for data farming.

In Swipe-AI the data is pre-labelled by an AI. The app presents the user the pre-labeled object along with a reference sample of what the AI believes the object is. This dramatically simplifies the user interaction for the human validators. Instead of trying to draw boxes on a smartphone the user can now simply swipe on the screen to accept or reject an AI proposal. By curating the dataset the validator then gets rewarded with data-tokens.

Goal of this project is to deliver a full MVP using this novel Swipe-AI approach.

All user interaction and reward functionality will be integrated into the DataUnion.app mobile version. The app will allow users with no prior crypto exposure to earn datatoken rewards. Evotegra will deliver the software to pre-label the data and the data itself taken from a test set. Using a test set allows us to measure the performance in order to optimize the process. The complex 274 class data set covers all kinds of traffic signs and lights in German traffic situations. A description of the algorithm can be found on the Ocean marketplace where Evotegra is selling the data to train it.

What problem is your project solving?

  1. Enabling AI and Human mass collaboration in order to curate high quality data in complex datasets.

Using a test set is important because it lets us evaluate the accuracy of the outputs of this novel method via the ground truth available from Evotegra. If this method is successful it can then be easily scaled to any other Computer Vision algorithms via the created infrastructure. There is no other product on the market that has a comparable flexibility as well as potential quality and throughput. .

  1. Enable potentially any smartphone user to earn rewards in data tokens

The app will enable onboarding and outreach to non blockchain mobile phone users. The app will not require any knowledge of blockchain technology to be used as the wallet of the user is embedded in the app and the claim/staking of the QUICRA-0 tokens that are the reward for this challenge is also simplified. This enables us to iterate on the contributor inflow and management.

  1. Enable highly flexible processes that can be easily applied to any custom high complexity dataset.

The creation of a DataUnion.app mobile version will be extended by more functionalities to contribute e.g. sourcing, manual annotation or verification of image data.

  • What is the final product (e.g. App, URL, Medium, etc)?

A mobile app for IOS and Android - it might be that Apple does not like the crypto angle of the app so we will have to use Testflight (an app testing app) to have IOS users participate.

  • How does this project drive value to the Ocean ecosystem?
  1. Exposing Ocean to any smartphone user worldwide
  2. By creating new business models it is the kickoff to an Ocean ecosystem, it will inspire others to find their Ocean based business model.
  3. Turning contributors into shareholder of the dataset generates new incentives related to quality
  4. Enabling a wide range of potential smartphone users worldwide to create an extra source of income
  • ROI

There are billions of potential contributors that will get into contact with datatokens and Ocean Protocol via this project. If 0.001% of them decides to invest in Ocean Tokens because of this project and we assume an average invest of 1000 $OCEAN, then this results in a potential of a TVL of * 0.001% * 1000 $OCEAN = 7.000.000 $OCEAN demand.

bang = 7.000.000 $OCEAN

buck = 6.000 $OCEAN

(% chance of success) = 80%

ROI = 7.000.000 $OCEAN / 6.000 $OCEAN * 0.8 = 1166 which is above the demanded ROI of 1.0

Project Deliverables - Category

IF: Build / improve applications or integration to Ocean, then:

  • App will be live, at: iOS AppStore + GooglePlay Store (url or app store)

  • App will be open-source with a permissive license at: github.com/DataUnion-app

Project Deliverables - Roadmap

  • Any prior work completed thus far?

A first prototype of the app is in the works as we have to start the admission process for the iOS app store as soon as possible.

Evotegra has the algorithm available as a deliverable that DataUnion.app can use to create the necessary input for the app and the data needed to evaluate the result of the crowdsourcing verification.

  • What is the project roadmap? That is: what are key milestones, and the target date for each milestone.

The app is planned to be released in April, due to the inability to predict the app store processing time it might have to be delayed to May but this will be communicated and the process will be documented.
As always in novel, complex software development there can be unexpected delays that might require a later release.

  • Please include the milestone: publish an article/tutorial explaining your project as part of the grant (eg medium, etc).

We will create a joined release blog post for this project.

  • Please include the team’s future plans and intentions.

  • Any maintenance?

We will extend this functionality in the future as a success in this project enables us to use our platform to verify other Computer Vision algorithms which we already have in our pipelines.

Project Details

If the project includes software:

  • Are there any mockups or designs to date?

In progress.

  • An overview of the technology stack?

Mobile App: React Native

Backend: Flask + Python libraries, CouchDB (+ PouchDB), Docker

If the project includes community engagement:

  • Running the campaign on social media for how many weeks?

We will actively promote the challenge with datatoken incentives for participants.

Team members

For each team member, give their name, role and background such as the following.

The team members are listed in the order of them joining the project.


Tobias Manthey

  • CEO & founder Evotegra, Regional Manager German AI Association, Teamlead AI, Software Developer

  • Twitter: @evotegra

  • LinkedIn Tobias Manthey

  • Website: Evotegra

  • 3Box Profile:3Box


Additional Information

Any additional information, custom fields, or images you would like to add? For example:

Market situation?

Currently there are no end-to-end solutions for the capturing, annotation and machine learning training available on the market. Especially not on this scale and with the involvement of worldwide contributors.


What’s the status on this?

Please check our video update for round 7 - in the mobile app part we show the current state of this proposal:

The data is now extracted, annotated, the communication with the backend works but there is still a bug that prevents this from working perfectly as intended. So we intend to squish that bug during July 2021.