Visual identification fails

[MadAlex21]

Dear Trex developer,
I am a student in the Ache lab at the University of Wuerzburg and I want to use Trex for fly tracking. I encountered a problem during the visual identification process.
When I start the visual identification in Trex the training process fails. It cannot define the function „learn_static::estimate_uniqueness“ because „learn_static“ throws following runtime exception:
ImportError: cannot import name ‚keras‘ from ‚tensorflow‘ (unknown location)
I suggest there are dependency-problems because I followed the installation-instructionsfor Windows, installed everything with conda, but had to install a new version of tensorflow via pip to start the training process. Now I’m having tensorflow (and keras) 2.12.0 (as it is used when installing Trex on Mac with M1 chip).
I’m happy about every suggestion how to solve this problem.

Computer:

• Operating system: Windows 10
• Build hash: 19044.2965
• Terminal used: Anaconda Prompt

[mooch443]

Hey, I realize this is late but - seems like your conda environment is broken? It says it cannot import keras from tensorflow. You could try out a beta version (be careful though) that uses all-together different versions of everything:

# Linux / Windows / macOS
conda create -n beta -c trex-beta -c pytorch -c nvidia trex

[tettcarolyn]

Hi Tristan! I'm having a similar issue running visual identification on the Trex Beta version using Windows 11. I was able to successfully convert the video, but when starting visual identification, this is the error that occurs:

[mooch443]

Hey there, unfortunately this does not really help me much in debugging this :( What version are you running? And it seems you are running it in cmd.exe. You should launch it from an anaconda powershell prompt, which is what I am doing on Windows usually. If you can, provide the entire log of running the program and try updating your version, too.