AI-PLATFORM FOR AUTOMATED TRAINING OF OBJECT DETECTION MODELS BASED ON CAD DATA
- /
- POC
- /
- AI-PLATFORM FOR AUTOMATED TRAINING...
Deep learning object detection is increasingly used in industrial applications, but its training requires a huge number of images and the objects that can be recognized are limited. This experiment aims to develop an automatic training procedure for AI object detection on large industrial datasets.
Start date: 01/06/2021
Duration in months: 18
Problem Description
Due to the time-consuming and cost of vision-based part recognition based on Deep Learning, only a few specific industrial applications. Applications such as warehouse part detection are still not economically feasible. Kimoknow aims to speed up these applications to offer them to a broader market.
Goals
New services
Challenges
Deep learning object detection is increasingly used in industrial vision applications, but its training requires a large number of images. This experiment aimed to develop an automatic training procedure for AI object detection on large industrial datasets.
Innovation results
The approach for automatic AI training entails generating synthetic images from 3D CAD and then using these datasets to train an object recognition model. Both stages have been accelerated using multiple GPUs. The computing time has been reduced from more than 100 hours to just 10 minutes.
Business impact
Production and logistics companies such as Gabler Engineering benefit from Kimoknow services for efficient object detection models, reducing manual effort and cost. This technology also enables new companies to quickly access AI-object detection results.