hsp-iit / ergocub-perception

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

ErgoCub-Visual-Perception

Installation

Environment

The best way to manage the environment is to use Docker. Build the image with:

cd docker
docker build -t ar0s/ergocub-perception-image .

TensorRT engine creation

This application contains several machine learning modules that are accelerated with TensorRT. Since the TensorRT engines are hardware-specific, it is necessary to create them everytime a new machine is used. Each TensorRT engine is built from an ONNX file. You can download the ONNX files from here, extract the "onnxs" folder to the root directory of the project. To build the engines, just launch:

./build_engines.sh

and wait untill each engine is built (last line should be "[I] Finished engine building in 2.615 seconds" or "Inference succedded!"). Run

./stop.sh

to remove the container and the tmux session. You can then erase the "onnxs" folder in the root directory.

Other files

Other files are needed, you can find them here.

Test installation

To test the installation, launch:

./start.sh -ys

Run with webcam

In order to test the pipeline with the webcam, follow ..

About


Languages

Language:Python 96.0%Language:Shell 3.1%Language:Dockerfile 0.9%