How to Train a Defect Detection Model Using Synthetic Data with NVIDIA Omniverse Replicator

Technical Blogs & Events Technical Blog
1

Originally published at: https://developer.nvidia.com/blog/how-to-train-a-defect-detection-model-using-synthetic-data-with-nvidia-omniverse-replicator/

Learn how to train an object detection model entirely with synthetic data, improve its accuracy with limited ground truth real data, and validate it against images that model has never seen before.

2

Thanks for the article, it’s a good reference design.

I started playing with Replicator using a similar method, i.e. putting all my Replicator code in an extension so I could use VS Code to debug. The problem I have though is Omniverse Code crashes after a few iterations through the debugger. I have a VERY simple Replicator extension and I’ve tried everything I could to free memory and such to avoid memory issues, but it still crashes. Headless mode works fine but there’s no debugger and I’m not the best Python programmer so I rely on the VS Code debugger / Github Copilot to hold my hand. It’s extremely frustrating when I have to restart OV Code every few tries of my Replicator code, obviously.

I’m on Windows 11 with 64 GB RAM and an RTX 4090. What kind of machine are people using to successfully work with Replicator extensions? What machine was this extension developed on. Should I just max out my RAM and pray or what? Should I switch to Linux? What’s a known reference machine specs for using Replicator with no crashes, currently?

Any pointers appreciated.

Thanks,
Dave

3

Your computer specs look great, I would start by making sure that OV code, your computer, and especially your GPU drivers are all up to date.

Let me know how that goes and we’ll move forward from there!

Eric