Object Detection Using YOLOv3/TinyYOLOv3 in Keras with a ROS Wrapper

This package contains 3 scripts with the following publishers and subscribers:

Subscribers:

• [topic]/image_raw: Image data from camera node.

Publishers:

• object_detector: Publishes label and prediction of the class with maximum score and coordiates of the bounding box, under the message type Predictor.msg
• object_detector/image_raw: Publishes image data with bounding boxes, labels and scores.

Note: classify_tinyyolov3_lite.py does not publish bounding box and image information.

Installation:

Dependencies:

• Install Tensorflow version 1.15 (requires CUDA 10.0)

  $ pip install --user --upgrade tensorflow-gpu=1.15

  If not training and facing troubles with CUDA installation, then install the CPU only version.

  $ pip install --user --upgrade tensorflow=1.15

  Ensure the system python path still points to python2.7

• Install Keras:

  Install Keras version 2.3.1

  $ pip install keras==2.3.1

  Install h5py. You will need this to load the models.

  $ pip install h5py

• Numpy:

  $ pip install numpy

• OpenCV

Model files:

Download pre-trained models (.h5) for TinyYOLOv3 and YOLOv3 trained on the MSCOCO dataset.

https://drive.google.com/drive/folders/1ni9L2r2xKmSfoaX6T8I2Biy53taakAn-?usp=sharing

Copy the model files into the models folder.