Crunchfish Gesture Interaction showcase image domain transfer tool

Crunchfish is proud to showcase its latest software tool for deep learning that enables the use of synthetic image data to generate annotations on real images – image domain transfer. Examples of image domains are different camera sensors such as color, greyscale and Infrared cameras. The new tool is part of a comprehensive image data management system that includes all steps from creating synthetic image data to transfer of real-life images for deep learning.

En bild som visar inomhus, person

Automatiskt genererad beskrivning

 

A crucial part of Crunchfish Skeleton Platform – enabling hand tracking and gesture interaction – is to create training data for deep learning and neural networks. All training data must be annotated. With millions of images, manual annotation is extremely time consuming why, Crunchfish has developed a tool that automatically generates annotated images with high accuracy at a tiny fraction of the time compared to manual annotation. This radically shortens the tedious annotation process.

 

Reliable and robust image data is essential for deep learning, which is why a huge number of high-quality images are needed. This data includes both image streams of real hands as well as synthetic data. By doing a photogrammetry scan of a hand, a 3D model is created and used to generate synthetic image data. The hand objects in this data can then be manipulated by adding animated anatomy and by modifying loads of parameters such as size, age, color, shadows, lighting effects, etc. In this way, a huge variety of images are created to secure a diversity of data content.

 

Crunchfish’s image domain transfer tool generates data for deep learning in a new domain – such as Infrared cameras – without any manual work, using only synthetic data. Changes and addition of new training data can be done in days rather than months which makes the versatile Crunchfish Skeleton platform much more efficient.

 

“With this new image domain transfer tool, we can increase the speed of generating new data substantially for new domains such as different camera types and enable us to bring new products faster to the market. Unsupervised domain transfer is a holy grail of deep learning and we can now reap the benefits of it”, says Daniel Milesson, R&D Director, Crunchfish Gesture Interaction.

 

To learn more about Crunchfish’s new Domain transfer tool, please join Crunchfish’s Survival of the fittest webinar air on Friday April 8th at 08:00 am CET / 11.30 IST. The webinar is broadcasted live and recorded, starting with a short presentation, followed by moderator led Q&A with a panel and the audience.

 

Register for the webinar on LinkedIn using the links below.

Gesture Interaction training data webinar 22#14

 

For more information, please contact:

 

Joakim Nydemark, CEO of Crunchfish Gesture Interaction AB
+46 706 35 16 09
joakim.nydemark@crunchfish.com

 

This information was provided by the contact person above for publication on 7 April 2022 at 08:30 CET.

 

Västra Hamnen Corporate Finance AB is the Certified Adviser. Email: ca@vhcorp.se. Telephone +46 40 200 250.

 

About Crunchfish – crunchfish.com

Crunchfish is a deep tech company developing a Digital Cash platform for Banks, Payment Services and CBDC implementations and Gesture Interaction technology for AR/VR, automotive and digital interfaces. Crunchfish is listed on Nasdaq First North Growth Market since 2016, with headquarters in Malmö, Sweden and with representation in India.

 

Download as PDF