Data on the Edge with Dataloop: Key Takeaways

December 15, 2021

In the second episode of Data on the Edge, Jeff Mills, Chief Revenue Officer of iMerit, sits down with Co-Founder and CEO of Dataloop, Eran Shlomo to explore how Dataloop, an end-to-end cloud-based annotation platform, helps to tackle edge case situations in computer vision and  what the future of the labeling tooling space looks like.

iMerit has partnered with Dataloop specifically on computer vision projects and together they have been able to customize workflows quickly for more than 50 iMerit projects to date.

Here are 5 key takeaways from the session:

  • The three pillars of Dataloop include – Data development, Human-in-the-loop as a core part of the process, feedback loop between humans and machines. Dataloop has two important roles in adding value to the labeling workflows: 
    • Effectively generate sustainable training sets with people 
    • Create a single cloud that is fast, smart and cheap
  • When a bug is reported on the Dataloop platform, the tool captures the information, analyzes it, and automatically recommends other files with similar quality issues to delivery managers and quality controllers, making the workflow more efficient and faster.
  • Dataloop is agnostic to file types including image, video, and audio. iMerit and Dataloop are also working extensively on workflow optimization for 3D images.
  • Dataloop is working on an application development SDK where one can spin out a full labeling application within hours. It will help assemble the right tool to the right job and vertical.
  • Major changes that can happen in the tooling space in the coming years:
    • Transition from batch to continuous or stream processing, and with advancements in processing power and neural networks, the ability to learn from humans in hours rather than months.
    • Differentiation of the debug and release paradigm for data
    • Self supervision