keras-yolo3

YOLOv3 implementation in Keras with TensorFlow backend trained to detect micromobility vehicles on datasets collected from LGSVL Automotive Simulator.

Introduction

The original implementation comes from qqwweee/keras-yolo3.

Our YOLO3 detects 5 different micro-mobility vehicles from the images captured by the LGSVL Automotive Simulator Micromobility version. The vehicles we are interested in detecting include scooters, hoverboards, skateboards, segways, and one-wheels.

Quick Start

1. Download pretrained weights here.
2. Move the downloaded .h5 file inside model_data
3. Run YOLO detection.

python inference.py [OPTIONS...] --image, for image detection mode

Usage

Use --help to see usage of inference.py:

usage: inference.py [-h] [--model MODEL] [--anchors ANCHORS]
                     [--classes CLASSES] [--gpu_num GPU_NUM] [--image]

required arguments:
  --image            Image detection mode

optional arguments:
  -h, --help         show this help message and exit
  --model MODEL      path to model weight file, default model_data/yolo.h5
  --anchors ANCHORS  path to anchor definitions, default
                     model_data/yolo_anchors.txt
  --classes CLASSES  path to class definitions, default
                     model_data/coco_classes.txt
  --gpu_num GPU_NUM  Number of GPU to use, default 1

ROS

If you are trying to run inferences on LGSVL simulators using ROS nodes, get_detections function in yolo.py returns 2D bounding box information.

ROS nodes that were created for this project are available here.

Training

1. Generate your own annotation file and class names file.
   One row for one image;
   Row format: image_file_path box1 box2 ... boxN;
   Box format: x_min,y_min,x_max,y_max,class_id (no space).
   For VOC dataset, try python voc_annotation.py
   Here is an example:

   path/to/img1.jpg 50,100,150,200,0 30,50,200,120,3
   path/to/img2.jpg 120,300,250,600,2
   ...
   

2. Modify train.py and start training.
   python train.py
   Use your trained weights or checkpoint weights with command line option --model model_file when using inference.py Remember to modify class path or anchor path, with --classes class_file and --anchors anchor_file.

3. You can access to our datasets we trained our model on from here. These two datasets are separated because they were collected in different times. Setups and formats are identical.

• Dataset 1

• Dataset 2

Results

Images

Video - Real-time inference on frames captured from the Simulator

Contributors

• Deepak Talwar - (https://github.com/deepaktalwardt)
• Seung Won Lee - (https://github.com/swdev1202)

This repository was originally developed for CMPE 297 - Autonomous Driving course at San Jose State University in Spring 2019.