ZED_Depth_Viewer v4.2.3 - Corrupted data?

Software Python
,
1

Hi,

I am trying to replay SVOs through the depth viewer. The SVO were obtained with v4.2.3 and the tools are running on v4.2.3.

I can easily visualize the images through the ZED_Explorer but I have no success with the ZED_Depth_Viewer.

It logs some error while opening my SVO files and the data seems corrupted:

in bool ImageHandler::initialize(sl::Mat&) : Err [999]: unknown error.
in bool ImageHandler::initialize(sl::Mat&) : Err [400]: invalid resource handle.
in bool ImageHandler::initialize(sl::Mat&) : Err [400]: invalid resource handle.
in bool ImageHandler::initialize(sl::Mat&) : Err [400]: invalid resource handle.
in bool ImageHandler::initialize(sl::Mat&) : Err [999]: unknown error.
in bool ImageHandler::initialize(sl::Mat&) : Err [400]: invalid resource handle.
in bool ImageHandler::initialize(sl::Mat&) : Err [400]: invalid resource handle.
in bool ImageHandler::initialize(sl::Mat&) : Err [400]: invalid resource handle.

image
image2271×1194 60.7 KB

Any idea on what could be the reason? We are running the tools through a nvidia-enabled docker image.

Best,
Noé

2

Digging a bit more into that. It seems to be an issue with the Depth Viewer, not the SVOs. Whether on 4.2 or 4.1, whether using a SVO or a live camera feed, the depth viewer is encountering the issue.

3

This looks like a CUDA issue, if you’re using docker make sure all GPU features are enabled, like display acceleration, hardware decoding, etc (GitHub - stereolabs/zed-docker: Docker images for the ZED SDK).
You can test simpler configurations like a CUDA sample or a ZED SDK tutorial (without display) to narrow down the problematic part

4

Thanks for your answer. The docker image I am using is already GPU enabled with all NVIDIA drivers capabilities. It also mounts the standard files required for X forwarding.

Thanks for reminding me about the Dockerfile samples. I’ll give a try to run the tools through the official image and report back.

I see there are no gl Image for 4.2. Could you please confirm whether the devel image is compatible with the tools?

5

Also, to clarify, I can work with the cameras using the SDK with my docker image. The only executable that is failing is the ZED_Depth_viewer

6

The tools do not seem to be built into the recent devel docker images. They can’t be found under /usr/local/zed

7

For future users, setting Performance Mode in Nvidia Profile fixed the issue.

8

This means that the AI models installed on your device are broken.
I recommend you run this command to clear them:
ZED_Diagnostic -aic

9

For future debugging, how could I reach this conclusion? Does the depth viewer interact differently with the models compared to the SDK?

The depth model was also working fine through the SDK.
I couldn’t display the images in the depth viewer, while they properly displayed in the explorer.

10

No, normally if you face issues by using the NEURAL depth mode in ZED Depth Viewer you should expect to experience the same problem with other applications using the same depth mode.

In any case, your problem seems to be not related to depth mode, but to image rectification. Indeed, ZED Explorer, which does not display rectified images, worked correctly.
Did you perform some kind of manual calibration?