scikit-image not installed through pip + tensorflow version for GPU support

[b-grimaud]

Hi,

Just as a heads up, if someone ever encounters those issues during installation :

• I noticed that installing deeptrack through pip in a fresh conda environment does not seem to install scikit-image by default, despite it being in requirements.txt. Installing it independently through pip works just fine.
• Whenever setting up an environment for GPU support, installing CUDA separately seems to break tensorflow, if it was installed as a dependecy of deeptrack beforehand. The solution that worked for me is to install tensorflow-gpu on a clean environment, and then install deeptrack.

[BenjaminMidtvedt]

Hi! Great heads up. scikit-image seems to have inadvertently been removed from the dependencies. This should be remedied in the next release! For the second point, I have not personally encountered this problem, but I also don't personally use conda. However, gpu-accelerated tensorflow is often a hassle to set up correctly, so it does not surprise me that there are cases where conda fails. I will try to reproduce!