Ahmed Aly

I think, and I can be completely wrong here, that Keras doesn't actually go ahead and make the network UNLESS you compile it.

Think about how the algorithm is evolving the weights. This will tell you how to implement the same algorithm to OpenAI Gym environment.

I agree, this is a requested feature because I CAN use Kite in JupyterLab and Jupyter Notebooks (not sure about the latter but should work!). So I expected that VS...

That would be helpful indeed, any luck?

Have you solved this? Just curious, I haven't worked with this lib before.

I see. The other libraries usually operate with TF 1.X. The advantage of this library is that it uses Keras, so the interface doesn't change much between TF1.X and TF2.X....

@nscotto can you tell me what other libraries worked for you? Did you measure the accuracy using COCO ?

What other SOTA pose estimation methods are you referring to? For me, the code is very very slow to start up and do the pre-processing steps.

@hnuzhy you can turn off the refinement part since it doesn't affect accuracy that much and is really slow.

Hello, I am facing the same issue, though I am using Petr on its own (not through PHO). Any ideas?