Home

Overview

This repository provides several nodes for running neural networks on ROS, including debugging and visualization.

caffe_ros

The caffe_ros node provides support for DNN inference using TensorRT. This node currently supports Caffe models such as classification (TrailNet), YOLO-based object detection, regression and semantic segmentation. The node produces messages in different formats depending on the type of the network and post-processing options (e.g. TrailNet vs YOLO).

The timestamp in the header of the message is the same as the timestamp from the corresponding camera message. This simplifies synchronization of the image frames with DNN output and other nodes for the purpose of debugging and logging.

image_pub

This node is used to publish an image or video onto a ROS image topic for debugging purposes.

output_view

This node subscribes to the output of caffe_ros and provides visualization (currently only segmentation is supported). For the node to work properly, caffe_ros and output_view must be set to the same post processing type (e.g. "Segmentation").