This repository contains the Palouse Robosub-specific ROS nodes for employing deep learning neural nets for vision inferences.
Networks must be stored at /repositories/robosub_object_detection/trained_models. To run specific network, run
roslaunch robosub_object_detection tensorflow.launch model:={name_of_network}
All networks should be available on the robosub server (robosub.eecs.wsu.edu)
at /data/vision/trained_models. To pull down new models, use
tools/pull_trained_models.sh.
The current implementation relies on TensorFlow 1.5 - this can be installed
through Pip. You will also need atleast version 2.6 of the proto-compiler
(sudo apt-get install proto-compiler). Please verify the proto-compiler
version before continuing.
Next, run the tools/initial_setup.sh utility to configure your repository.