/YOLO_3D

ROS Package for 3D Object Pose Estimation

Primary LanguagePythonMIT LicenseMIT

YOLO 3D

ROS Package for Estimating 3D pose of an object using YOLOv3 Tiny from RGB + Depth Image.

** The pakcage has been developed and tested on ROS Melodic.

Configuring

  1. Install the required Libraries
$ pip install -r requirements.txt

  1. Place the YOLO weights and config file in yolo folder and change the path accordingly in init function of Perception class.

  2. Change the image topics in launch file.

  3. Make the demo file executable

$ chmod +x demo.py
  1. Execute the demo file
$ roslaunch YOLO_3D estimate_pose.launch

Hardware Requirements

  1. Depth Camera

Training your custom YOLO Model

The current weights are trained on few objects from the YCB dataset. If you wish to train the model with custom dataset, follow this amazing blog.

** The Jupyter Notebook provided by the blog author throw errors due to newer version of darknet. You can follow this notebook for training - Notebook

Authors

Gaurav Sethia, Siddharth Ghodasara, Kaushik Balasundar

References

  1. P. Adarsh, P. Rathi and M. Kumar, "YOLO v3-Tiny: Object Detection and Recognition using one stage improved model," 2020 6th International Conference on Advanced Computing and Communication Systems (ICACCS), 2020.

  2. http://wiki.ros.org/image_geometry

  3. https://medium.com/@today.rafi/train-your-own-tiny-yolo-v3-on-google-colaboratory-with-the-custom-dataset-2e35db02bf8f