Object Detection with YOLOv2

Final project by Group 2 (Localization) in HDU CS Innovation & Practice Course (2018 Spring).

Inspiration

This repo is based on Car-detection-PA, containing a tutorial for YOLOv2, which is a programming assignment in Convolutional Neural Networks by Andrew Ng.

Requirements

Keras
Tensorflow (For Keras backend.)
OpenCV
h5py (For Keras model serialization.)
Python 3

Installation

Use pip to install the required packages in your command line. Visit the official sites above for more details.

pip install numpy h5py
pip install tensorflow # CPU-only
pip install keras
pip install opencv-python

Get Started

Download (Password: 8kim) an existing pretrained Keras YOLO model stored in yolo.h5. (These weights come from the official YOLO website, and were converted using a function written in YAD2K) and put it into model_data/ folder.
Put your test image/video in images/ or videos/.
Set your image/video file and shape in yolo_v2.py.

if __name__ == '__main__':
    '''
    code
    '''
    # Set the original image/video shape
    image_shape = (960., 544.) # (height, width)
    
    # Detect a video
    video_file = "traffic.mp4"
    predict_video(sess, video_file) # output in out/

    # Uncomment the code below to detect an image
    # out_scores, out_boxes, out_classes = predict_image(sess, "person.jpg")

Run the model in your command line.

python yolo_v2.py

The prediction info will be printed into output.txt.
The notebook version will be uploaded very soon.

More Details

# todo

TODOs

Unify the interface in the video detection part.
Train the model.
Try YOLOv3.
...

Reference

Joseph Redmon, Santosh Divvala, Ross Girshick, Ali Farhadi - You Only Look Once: Unified, Real-Time Object Detection (2015)
Joseph Redmon, Ali Farhadi - YOLO9000: Better, Faster, Stronger (2016)
Allan Zelener - YAD2K: Yet Another Darknet 2 Keras
The official YOLO website (https://pjreddie.com/darknet/yolo/)

n3rdd / Object-Detection-YOLOv2