Skip to content
Switch branches/tags

Latest commit


Git stats


Failed to load latest commit information.
Latest commit message
Commit time

Decentralized AI (final project for Siraj's School of AI)


Benoit Courty, Matthew McAteer, Alexandre Moreau and Jeddi Mees

For more background info read our Whitepaper.

This is our try at building a "Decentralized AI". Well it is just a semantic segmentation task that run in a decentralized fashion : The task is done on a machine in the internet, like in a proprietary cloud, but on a decentralized cloud : you do not have to create an account with the computer owner. All is handle by iExec.

The semantic segmentation is done by the Mask RCNN project trained on the COCO Dataset.

Submit an image : FrontUI

Get the result in the work tab : FrontUIWork

And you are done : SampleResult

Other sample : MASK_R-CNN

The Docker image was based on the Modern Deep-learning container from Waleed Abdulla, with the Mask RCNN added into it along with a modified version of the demo packaged for iExec.

iExec is a whole ecosystem with a market-place for DApps, Oracle mechanism, scheduler, workers,... Dedicated to off-chain computing in a fully decentralized way.

The V2 is just out (speaking from 1st of June 2018).

iExec SDK is a NodeJS application who allow to easily create and manage your application.

The result is that you can call it quite like an API to get your resulting image :

How to Run

iExec Front Side

You could use it on the browser :

Get ETH and RLC for Kovan : . Connect to Metamask and switch to Kovan Ethereum test network. Ask for free ETH on Kovan faucet and for free RLC on iExec marketplace, then transfert RLC from your wallet to your "account" (on top left of iExec marketplace)

Build it from source :

cd frontend
npm install
npm run dev

Your browser will automatically go to localhost:8081 so you can access the frontend. . Choose an image from your harddisk or copy-past an url. . Choose a worker in the list on the right. . Click on "iExec" button.

OpenMined Side

In openmined folder you will find a demo of how to use Open Mined to train a model using decentralized grid computing capabilities.

How we make it

Building the Docker Image

docker build docker_keras_cpu/ -t trancept/keras_mrcnn:v0
docker run -v $(pwd):/iexec trancept/keras_mrcnn:v0
docker push trancept/keras_mrcnn:v0

iExec project

# Init project

# Get money
iexec wallet getRLC
=> For ETH, on Kovan you have to go to ask for it on [Kovan faucet](
# Check your wallet
iexec wallet show
=> You need to have ETH and RLC
# Send money to the iExec account/Marketplace to use it
iexec account deposit 100
# Check money
iexec account show


Adding docker image to iExec :

  • Edit iexec.js
  • Run :
iexec app deploy
iexec app show

Prepare order

iexec order init --buy

Important : You have to edit iexec.json at these step to edit the "params" string to match the parameters you want to send to the job.

How to execute iExec Dapp

Easiest way

The easiest way is to go to and place a Buy order with :

Command line way

. Clone the repository . Change the image url in iexec.json . run

You have to initiate an order to buy computing ressource, then find one available, then buy it !

# Show available computing ressource
iexec orderbook show --category 3
# Check a ressource
iexec order show 170
# Buy the ressource
iexec order fill 170
# Check the status
iexec work show 0xfda65e0d09bf434ea1e52f4ec044a07d6e7d592d --watch --download


AI computer vision on the Ethereum Blockchain with iExec




No releases published


No packages published