Carlos Andrés Álvarez Restrepo

Any updates on this? Same error

> @charlielito this is almost done right? Yeah. With the MNIST Example. But we were discussing maybe including a more complex example for TPU?

This happens with python2. @serbuh To solve that, import this at the beginning of the `__init__.py` and `pyfakewebcam.py`: ``` from __future__ import absolute_import ```

I adapted the `demo/demo_inference.py` to work with live camera. Here is the code ```python # Copyright (c) OpenMMLab. All rights reserved. from argparse import ArgumentParser import cv2 import numpy as...

Waiting for response...

@fujitatomoya my bad, actually, I was using Cyclone DDS `export RMW_IMPLEMENTATION=rmw_cyclonedds_cpp` After changing it to `rmw_fastrtps_cpp` I can confirm your results and the expected behaviour. In that case what should...

IMO this should be fixed since the behavior should be the same independently of the DDS implementation.

I mean that it is very confusing that it behaves differently without any warning because the user would expect the same behavior. I think the feature is also a reasonable...

> Pass verbose=False in the inference function It indeed stopped the logs, but now the masks are not being drawn. How can I stop the logs but still get the...

I am using it to do online predictions so I don't want to serialize the results as an image. With `model(img, verbose=True)` the masks are drawn in the image without...