NeuralNetwork training data visualisation

You work in the Data Management Team at Continental’s ADAS BU, and you support the supply chain of neural network training data. You received the following email chain. Help Sven with delivering the requested input for decision making. Make sure all relevant information in the emails are considered.

From: Mjölnir, Sven Christian <svenchristian.mjoelnir@>

Sent: Tuesday, March 19, 2019 3:23 PM

To: You

Cc: Bajzak, Dorottya Agnes <dorottyaagnes.bajzak@>; Ridzi, Peter Lorant <peterlorant.ridzi@>

Subject: RE: prelabelled data for categorization


I have attached the output of a rudimentary object detecting NN on the test recordings (with sampling per second, bounding box output, image size: 3584*1896, used classes: bike, bus, car, motor, person, truck). I also attached some samples projected on the images.

Please support Dorka in choosing which recordings – or even what parts of recordings – to send to TeamA and TeamB. Any sort of visualization is OK, the point is to make the decision as quick as possible. This is just the test round, but bear in mind, we will need to be able to replicate this process on 100+ recordings daily from Q2.

Thanks for your help,




From: Bajzak, Dorottya Agnes <dorottyaagnes.bajzak@>

Sent: Thursday, March 14, 2019 1:23 PM

To: Mjölnir, Sven Christian <svenchristian.mjoelnir@>

Cc: Ridzi, Peter Lorant <peterlorant.ridzi@>

Subject: prelabelled data for categorization

Hi Sven,

as promised, you can find today’s recordings in the orders folder. What we need is to find a way to quickly but correctly determine of any recording, for which projects it could be best used.

For this test, we made three recordings. Two annotation teams should get the proper materials: TeamA works on a VRU behaviour prediction project (vulnerable road user), TeamB works on an L3 highway autopilot project. Please provide any relevant information that is needed to sort out the content of the recording without looking.

Thank you for your support, looking forward to the results.



NN training data visualization decision making cooperation statistics


  • evaluate your colleagues’ needs and find key indicators – create concept
  • 2 points
  • filter necessary information on record-level
  • 2 points
  • show statistics of the content of the recordings (consider classes, timeline, etc)
  • 2 points
  • display information in proper layout for non-professionals
  • 3 points
  • make further recommendations (see submission last sentence)
  • 1 point


You can use anything, and you can make anything. The point is, that your economist colleague will be able to use and understand what you give to her, without any special competences. If you have any preliminary suggestions about the destination of each recording, or if you have a vision, how the large-scale process could look like, share your ideas.