Hi, I’m having an issue with optimizing my neural network. It doesn’t proceed at all, whether I try from the diagnostic application or directly from the command line, even after cleaning the models. When trying to optimize from the command line after downloading the model, it starts optimizing at 0.1% as shown in the image, but then it stops. I’ve tried using the latest SDK 4.25 and 4.1 with CUDA 12 and 11.8, and I’ve fully uninstalled and restarted everything. ZED_Diagnostic_Results.json (20.7 KB)
Hi @MechatronicMan
with RTX GPU cards we recommend using CUDA 12.
Please uninstall CUDA 11.8, verify that you installed the latest version of the GPU driver, and then install CUDA 12.6.
Then uninstall the ZED SDK and reboot.
Finally, install the ZED SDK v4.2.5 and verify with ZED Depth Viewer that you can correctly optimize the NEURAL Depth model.
Hi, I have problems with visual studio (camke generate files correct) but not specialist in and have problems with other librarys also… But i have other test on the same machine with win10 zed sdk 4.25 and cuda 12.6 also the same error ZED_Diagnostic_Results-win10 (8.5 KB)
After investigating and for future readers, the ZED SDK 4.2 doesn’t support Blackwell GPUs. We’re working on supporting it in a new release (it will require TRT 10 and CUDA 12.8)
I have the same problem with a RTX 5080 and can’t get the neural depth mode working (ZED SDK 4.2, CUDA 12.6). Are there any updates on the SDK to resolve this problem? Thanks
So the depth viewer and neural mode works well, but I just tried to use the python API and I have some problems with numpy incompatiblities. Is there a specific version required (I tried the version that was recommended for the old SDK and I tried numpy2.24).
" import pyzed.sl
Traceback (most recent call last):
File “”, line 1, in
File “pyzed\sl.pyx”, line 1, in init pyzed.sl
ValueError: numpy.dtype size changed, may indicate binary incompatibility. Expected 96 from C header, got 88 from PyObject"
Thank you
