Encountered Problems/Error #1
Comments
|
Hi Cedric, thank you so much for your feedback, its greatly appreciated. We're currently trying to reproduce your issues and incorporating your feedback into the next version of SIPEC, plus we will add some more detailed documentation and exemplary data. I will also document how those issues have been addressed Thanks a ton and hopefully hear soon, Markus |
damaggu
added a commit
that referenced
this issue
Jun 23, 2021
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Hi,
First I want to preface this post by stating that I'm quite new to Python and programming in general so my description of problems and the terms I use are probably lacking.
System used:
Linux 4.15.0-123-generic x86_64 with Ubuntu 18.04.5
My questions would be:
Are the template pipelines (segmentation.py etc.) complete (besides changing the paths) and useable for the full process from training with annotated data to actually using the trained neural network on a new video or is it expected that the user also makes changes in the code? Because from what I understood, the "segmentation.py", "behavior.py" etc. modules only train the networks and require annotated data. So which module would be used to analyze new videos with already trained networks? Is it the "full_inference.py" module?
I tried to run the "segmentation.py" module with the command
python ./SwissKnife/segmentation.py --operation train_mouseafter changing the following things in the code/system:
model.init_training(model_path=model_path, init_with="last")to
model.init_training(model_path=model_path, init_with="coco")Edit: I managed to solve the problem. I didn't read the code properly and did not store the frames in a "train" or "val" folder which lead to an empty frame variable.
\
Additional remarks for the installation process that might need to be changed:
installing PyYAML 5.1 from requirements.txt throws an error because it tries to uninstall the newer version first which was not possible because "it is a distutils installed project"
after installation through the requirements.txt anaconda gave a message that for some of the installed packages the dependencies are not correct. For example tensorflow 1.14.0 needs tensorboard >=1.14.0,<1.15.0 but in the requirements.txt it specifies tensorboard==1.10.0. There were more incorrect dependencies but I'm not sure which they were.
As mentioned above, the keras\application as well as the keras_application folder that was downloaded through pip-install did not contain a "resnet101.py" file and the respective "init.py" files have no "reference" to it. I also tried to pip install the newer keras-application package but it didn't help.
\
I'm sorry for the wall of text and hope that some of my described experiences can maybe be useful to ease the installation process and the use of this software for beginners like me.
Kind Regards
Cedric
The text was updated successfully, but these errors were encountered: