ingleShotDetector error - cannot import name 'set_guard_fail_hook' from 'torch._dynamo.eval_frame'

Question

Hello,

I am testing this sample scenarios: https://developers.arcgis.com/python/samples/automate-road-surface-investigation-using-deep-learning/ and when I got to point:

ssd = SingleShotDetector(data, backbone='resnet101',focal_loss=True)

I got the error:

ImportError: cannot import name 'set_guard_fail_hook' from 'torch._dynamo.eval_frame' (C:\Users\vholubec\AppData\Local\ESRI\conda\envs\CloneWars\Lib\site-packages	orch\_dynamo\eval_frame.py)

I am using Pro 3.3 with ArcGIS Deep Learning Essential package, so all should be fine - don´t you know if there is some workaround?

I found similar error here: https://github.com/unslothai/unsloth/issues/243, but nothing has helped to fix it in Pro.

Thank you.

PavanYadav · Accepted Answer

Hello @VHolubec Is it only the Single Shot Detector Model Type not working? Are you encountering issues with other Model Types (e.g. RetinaNet) or any other deep learning tools? This would help determine if it's a deep learning package installation issue.

VHolubec · Accepted Answer

Hello @PavanYadav,

after further steps - new clonned environment and installing DL Libraries from GitHUB from GitHUB, which became available a while after I asked, it has started to work.

So it look, there was some temporal issue, solved on the package side.

Thank you very much!