Latest update shows problem with recognition and reidentification models

[divdaisymuffin]

Hi @xwu2git,
We recently encountered few issues with latest update of Smart-City-Sample:

1. The entrance part does not work and shows error with Person-reidentification model.
2. The emotion recognition model when put in pipeline, also shows same error.
3. Both were working fine with 2-3 month old version.

For reference, I am adding screenshots of both.

[xwu2git]

Reproduced. Looks like a model mismatch to the latest OpenVINO version. Investigating.

[xwu2git]

@QuPengfei @nnshah1, Can you take a look? I tried to upgrade to different versions in the OpenVINO model zoo models. None of them works.

[divdaisymuffin]

any update about it?

[nnshah1]

@divdaisymuffin , @xwu2git

I took a quick look and this seems related to the INT8 versions of the model specifically. Downloading and converting the models with the latest public VAServing dependencies does seem to work but those same models do not work in the latest tip of openvisualcloud. I'll continue looking to see if a compatible INT8 model can be found, in the mean time from a functional perspective you can try removing the INT8 models temporarily and fallback to FP32.

[nnshah1]

@divdaisymuffin, @xwu2git I did some more digging and found the issue may be related to the model proc file and changes in DL Streamer. Can you check to see if this pull request fixes the issue for you: #783