This test uses the synthetic data generated in the previous test ("Pallet Jack" images in a logistics setting) to train a CV AI object detection model.
The AI development environment configurations are based on NVIDIA-Omniverse synthetic-data repository examples , which leverage the NVIDIA open-source Pre-Trained Tool Kit (TAO).
High-level training, validation, and real-world inference steps:
- Convert synthetic data metadata records into a suitable format for an AI model.
- Use an NVIDIA TAO detectnet-v2 container to convert KITTI to Tensorflow format.
- Download a pre-trained CV AI model (ResNet-18 model).
- Train the model on synthetic data.
- Validate model on synthetic data.
- Perform training and build validation model with synthetic data.
![This image shows an extract from the jupyter notebook from the NVIDIA synthetic-data-example repository](/static/media/9198938f-8c47-5a0e-82d9-6db6a62cd3f7/DAM-6e61b5f8-38f5-48c2-a31d-f85c6647cf47/out/image16=GUID-A0BA4D1A-A84D-4068-B24F-29AA50885262=1=en-us=Low.png)
Results
The provided juypter notebook contains the code required to run training, validation, and inferencing.