ESS DNN Stereo Disparity

Autonomous Machines Robotics - Isaac Isaac ROS
1

Hi

I am using ESS DNN Stereo Disparity model for my project. The model output is very noisy in case of less texture region.

left_image_00
left_image_001920×1080 375 KB

right_image_00
right_image_001920×1080 358 KB

disparity
disparity960×576 131 KB

I have uploaded the stereo images and its disparity map from the model. The camera is placed at a height of 700mm from the ground level.

When i try to convert the disparity to depth map, in PCD file data the noise is very high.
Is this the expected behaviour of the model ? Are there any limitation for the model in case of calculating disparity images?

Thanks in advance!

2

Hi @gayathri4,

I don’t think the Android sub-category is the correct area for this post. Since it is very specific regarding computer vision I think you should start in that category.

I took the liberty of moving this post over there.

I am sure someone will be able to assist!

Thanks!

3

I believe the Isaac ROS Team will be in a better position to offer assistance - so I have added the the tag and also moved this posting.

4

Hi @gayathri4 and thank you @nadeemm for redirecting here.

We made huge changes to Isaac ROS DNN Stereo Disparty with the new release of Isaac ROS 2.0, have you tried again?

New documentation: NVIDIA Isaac ROS — isaac_ros_docs documentation
Release notes: Release Notes — isaac_ros_docs documentation

Let me know,

Raffaello