search

MikeOfZen / Yet-Another-Openpose-Implementation

This project reimplements from scratch the OpenPose paper (Cao et al,2018), Using Tensorflow 2.1 and optional TPU powered training.
Mozilla Public License 2.0
92 stars 26 forks source link

Callbacks codeblock in Notebook not working #4

Open Neorath opened 4 years ago

Neorath commented 4 years ago

Hello! I'm trying to get this repository to work for a school project. I understand that you have developed this in tensorflow 2.1, however this version is no longer available and I'm forced to use tensorflow 2.2. I tried downloading the tensorflow 2.1 package from sourceforge and installing it using pip, but this doesn't seem to be working because a setup.py file is missing, and I don't know how to make one that would work. When trying to run the Callbacks codeblock in the Jupyter Notebook (after running all the ones above it), I believe the version difference is causing an error which is as follows:

`--------------------------------------------------------------------------- TypeError Traceback (most recent call last)

in 5 ] 6 if cfg.SAVE_CHECKPOINTS: ----> 7 checkpoint_callback,checkpoint_path=callbacks.make_checkpoint_callback(cfg,nowt,REAL_EPOCH_STEPS*cfg.BATCH_SIZE) 8 all_callbacks.append(checkpoint_callback) 9 TypeError: cannot unpack non-iterable ModelCheckpoint object` It seems to me that the return value may have changed between versions but I am not entirely sure about this. I have tried to remove the checkpoint_path variable declaration from line 7 after which it does run, but after doing so the Training part of the notebook doesn't work and I imagine this piece of the code is why. Do you have any advice on how to troubleshoot this issue? Thanks in advance!
Anqi-nus commented 3 years ago

same issue here! sorry it's been so long since the issue was posted, but have you found a solution to this issue?

Neorath commented 3 years ago

No sadly I didn't find any solution. Due to time constraints I had to give up trying to get this solution to work and instead I opted for the official OpenPose solution. I think if you can get tensorflow 2.1 working locally (you might have more experience or luck with getting packages to work locally), it should work just fine.

Anqi-nus commented 3 years ago

Thank you for the suggestion! I'm already using tensorflow 2.1 locally tho, it still face this error :( I'm unable to turn to the official solution due to their MATLAB dependency, luckily I could still get a working trained model without using the callback function.

bilalkhann16 commented 2 years ago

same issue.