SSDMobilenet2ONNX Tutorial

[rmccorm4]

Hi @guschmue ,

Was just glancing at https://github.com/onnx/tensorflow-onnx/blob/master/tutorials/ConvertingSSDMobilenetToONNX.ipynb, and had a quick question.

I noticed these lines in the first cell:

# force tf2onnx to cpu
os.environ['CUDA_VISIBLE_DEVICES'] = "0"

This line os.environ['CUDA_VISIBLE_DEVICES'] = "0" is actually specifying that CUDA applications will see/use only the first (zero-indexed) GPU on the system if any.

So the comment above that line is a bit misleading, and actually incorrect if you're expecting the CPU vs. GPU conversion to have different results, such as in this issue: #801 (comment)

Just wanted to point it out to avoid misleading people, unless I'm mistaken. 🙂

[guschmue]

oh, good catch! I wanted it to be "" and will fix it shortly.
For this specific tutorial it should not break something I think.

[guschmue]

should be fixed in master.

[rmccorm4]

Hi @guschmue ,

I see it's fixed in the efficientnet-lite tutorial, but still seems the same for the SSDMobilenet tutorial on master, that I originally brought up.

Looks like the latest commit for that tutorial was a no-op (just removed a new-line). Just mentioning in case it was accidentally missed.

Thanks,
Ryan

[guschmue]

Oh, sorry - I messed up. Let me try again.