Evotegra 100.000 image traffic dataset @ 1 Ocean

Name of Project: Create a professional 100k image traffic dataset

Proposal in one sentence

Create a professional traffic image dataset with at least 100.000 images with ~300 categories and an estimated million objects, which will be published on the Ocean Market with an initial data token price of 1 Ocean.

Description of the project and what problem is it solving:

Public datasets often contain low quality data. E.g. in the example below from the Apolloscape Dataset you can see that annotations contain objects invisible behind concrete walls. For any deep learning algorithm this is toxic data that will negatively affect the performance.
We propose to create a professional high quality traffic dataset of at least 100.000 images in ~300 classes, including traffic signs, cars, pedestrians, busses and several types of traffic lanes. We are going to collect that data from the scratch. It will contain data from highways, federal as well as country roads and cities to villages . We annotate that data that is going to have an estimated million annotated objects.
The goal of our proposal is to establish Ocean Market as a prime source for high quality data. Therefore we are going to publish this dataset on the Ocean Market in a pool with an initial data token price of 1 Ocean and least 2500 datatokens.

Category

Unleash data

Metric

Data Consume Volume

What is the final product?

The final product is an image dataset published in a AMM pool with an initial datatoken price of 1 Ocean and at least 2500 data tokens.

Grant Deliverables:

  1. Collect 100.000 images from diverse road scenes in Germany (~2-3 month)
  2. Annotate the images in ~300 classes (~1 month)
  3. Publish the dataset in a pool for the initial price of 1 Ocean (~September 2021)

Funding Requested:

17,600 USD

ROI

Our goal is to establish Ocean as a prime source for high quality data. The attractive entry price should encourage many first time users to get familiar with wallets, Ocean tokens, the Ocean Market.
We expect to achieve about 1 million annotated objects. While our proposal is coming at an estimated cost of 14k USD using a professional labelling company would cost about 50k USD plus the collection another 50-100k USD.

Proposal Wallet Address:

0x61B15998893cC746B46C08FEdEE13a0d1b33bBa9

Team

Web: www.evotegra.de
Email: manthey@evotegra.de
Twitter: @evotegra
Country of Residence: Germany

Previous OceanDAO Grant?

Yes

1 Like

A few notes here as I can’t change the proposal anymore.
The due date for the dataset would be about 3 month after grant. So October to November. The September is a copy paste error.
This proposal also contains outreach/marketing to actively make people aware of the dataset especially within the German AI association.

Absolutely supported for obvious reasons and close collaboration over a long time.

1 Like

Our image collection, annotation, and publishing process follows a waterfall-like structure.
Although we have 50% of the grant carried over, we are realizing the expenses for this batch of data now (Round 8), and amortizing the payment/work over the next 1-2 months.

Here is the work delivered & realized.
[X] Collect 100.000 images from diverse road scenes in Germany (Completed September 2021). This was completed within the budget and time. It cost 50% of the grant, and allowed us to kick off the annotation process.
[X] TBD: Annotate the images in ~300 classes (~1 month). Annotation can now start since data collection was completed. This will cost 47.5% of the remainder grant amount, and should be delivered by October to November.
[X] TBD: Publish the dataset in a pool for the initial price of 1 Ocean (~1 month). Publishing will take place after annotation is completed. This will cost a final 2.5% of the grant available.