Dependencies since 0.1
(Further to #4) 3 new dependencies have been added:
- AdaBins: https://github.com/shariqfarooq123/AdaBins
- MiDaS: https://github.com/isl-org/MiDaS
- pytorch3d-lite: https://github.com/MSFTserver/pytorch3d-lite
None of these are packaged so cannot be pip-installed :disappointed:
- AdaBins could be forked and made installable (depends on whether its weights are used alone? TODO: check how it's used)
- MiDaS seems to be available from torch hub and could perhaps be used in the same way it's used in the Gradio demo (see app.py there). However it's used as a package (
from midas.dpt_depth import DPTDepthModeland so on) - pytorch3d-lite is used as a package
py3d_tools
Essentially anything that is expected to be provided with a package name on the Python path should be forked and made available as a pip installable package under that name.
AdaBins has no package name, I'd call it adabins, so (after creating it) it'd be installed as
pip install -e git+https://github.com/lmmx/AdaBins.git@main#egg=adabins
MiDaS has the package name midas so would be installed as
pip install -e git+https://github.com/lmmx/midas.git@master#egg=midas
I've forked the pytorch3d-lite repo so it would be installed as
pip install -e git+https://github.com/lmmx/pytorch3d-lite.git@main#egg=py3d_tools
i ran into this while doing my own tard-refactor. can we just dump the adabin,MiDaS, Pyt3dlite gits into ours so that it is all cloned as one or is that against the rules?
Ha, oh hello :smile: I was about to just fork and package them in a git repo suitable for pip installation as outlined above, but yes you can incorporate as submodules, it just means you have to put more code in a single repo. I've seen that done but the idea of just making them distinct repos seems fine and neater (less lines of code in this repo)