ESS-DNN Stereo Disparity resources

we are currently testing out the ESS-DNN for some use cases and are very pleased with the results so far.
Looking forward to future usage - are there plans for NVIDIA to release the resources/scripts for training and exporting the model with TAO? Our camera inputs are typically much smaller than the current 576x960 input resolution, so a possibility to resize the model would be much appreciated by us. I think currently with the available ess.etlt, no resizing of the network is possible?

Is there a rough roadmap, of the further development of this net?


We’re glad DNN Stereo Disparity based on ESS is working out well for you! There are no plans to release the resources/scripts used for training and exporting the model at this time. We’ll consider adding the ability to resize the model for different resolutions or perhaps provide a model trained on smaller image sizes to our roadmap. Thanks for the feedback!

We’re continually working on improving the ESS model on NGC but we tend to publish major updates alongside Isaac ROS releases every few months.

1 Like

Thank you for your quick response!
I’m looking forward to that.