PalouseRobosub Object Detection

This repository contains the Palouse Robosub-specific ROS nodes for employing deep learning neural nets for vision inferences.

To run with specific network run

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}

To pull trained networks

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.

Requirements

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.