Missing PatchPerPix.models

[steriegler]

Hi everyone,

this is rather related to the archived PatchPerPix_experiments since I'm trying to replicate the wormbodies experiment. Ultimately, I would like to use PatchPerPix on my own 2D data. I have, however, not yet managed to reproduce the experiment. While I believe the dataset is arranged correctly (I used consolidate_data.py & trainTestSplit.py), I'm not even getting that far.

In the step mknet it appears that PatchPerPix.models, and therefore unet.py is missing. This is the case for both, PatchPerPix_experiments and PatchPerPix. I tried to get around that by using the UNET from https://github.com/Kainmueller-Lab/funlib.learn.tensorflow.git@a521614484423e3118ce024758d350b17f81b84d#egg=funlib.learn.tensorflow. This one is unfortunately not very compatible with mknet.py, and so far I haven't managed to find a unet.py that is.

Hence I would like to ask whether you could deposit a complete setup with all scripts necessary to run the wormbodies experiment successfully. Additonally, I was wondering whether an example config file with annotations of all parameters is available.

Thank you very much in advance! Loving what you achieved here, I'm looking forward to reproducing it!

[abred]

Hi Stefan, very sorry about the late reply! thank you for your interest.

PatchPerPix.models should be part of the PatchPerPix repo (https://github.com/Kainmueller-Lab/PatchPerPix/blob/ea4e2d4fb824d46e74c514ab8c7946df3467b652/PatchPerPix/models/init.py), did you follow the installation instructions in the readme file for the eccv2020 tag/commit? Can you please check if you can import PatchPerPix or PatchPerPix.models?

[tP1rx]

The lack of PPP.models also bothered me. The original keras implementation was rewritten in pytorch, so I assume the structure was reworked as well. The readme describes the pkg structure. The core of the current master branch is located in experiments/flylightsetups/setup. Models are stored in torch_model.py