This repository was forked from the original tensorflow/models so that we can make various research models pip-installable and pin their versions.
Each model that we want to support is on a different branch that includes a version tag (e.g. deeplab-0.0.1). Whenever we want to pull in upstream changes to a model, those changes should be pulled into a new branch with the version tag incremented.
Making a model pip-installable is just a matter of adding a setup.py file into its corresponding directory. See research/deeplab/setup.py and research/object_detection/setup.py for examples. A model's setup.py should only be added to its corresponding branch(es) in order to prevent accidentally installing a model from the wrong branch.
Once a valid setup.py exists, the model can be installed using the following syntax (using deeplab as an example):
pip install git+https://github.com/autognc/models@deeplab-0.0.1#egg=deeplab&subdirectory=research/deeplab
This syntax will also work as a dependency in requirements.txt or environment.yml files.