I am working on a 3D tracking project using Sparse4D (RN101) in the current TAO 6.25 ecosystem. I have successfully exported the model to ONNX, but I am stuck on the next steps for the TensorRT engine generation on Jetson Thor, as it appears to not be present in the current TAO version.
Instead, the documentation for deploying Sparse4D leads to the Halos-Outside-In-Safety EA page, which still lists a Q2 2025 availability. Since we are now in Q1 2026, has this deployment logic been made available to accepted applicants for the program? My application is currently undergoing review.
Any clarity on the roadmap would be greatly appreciated.
The TAO doc says as follows.
Refer to the Nvidia Spatial AI documentation page for more information about deploying a Sparse4D model to DeepStream via TensorRT engine generation..
Could you please open below link to run? https://developer.nvidia.com/docs/oiss/spatial-ai/3D-multi-camera-detection-and-tracking-Sparse4D.html#guide-integrating-new-sparse4d-model
I opened the link and as before it leads to a page mentioning I need access to the Halos-Outside-In-Safety program, and I’m currently waiting for access approval.
Despite the page mentioning a notification once the software is available, I suppose it already is available and I just need to wait to be granted access?
Yes, after granting access, the https://developer.nvidia.com/docs/oiss/spatial-ai/3D-multi-camera-detection-and-tracking-Sparse4D.html#guide-integrating-new-sparse4d-model should be available to open.