ChelovekHe / yolov2

YOLOv2 Implementation in Keras

YOLOv2 - Object Detection Model on Keras

---

This repo is the implementation of YOLOv2, an object detector using Deep Learning, discussed in "YOLO9000: Better, Faster, Stronger"

Project Status: Under Development!!

Dependencies

• Set up environment

conda env create -f environment.yml

• Activate environment

source activate yolo

• Install OpenCV

conda install -c menpo opencv=2.4.11

Usage

• Download weight files and place under weights directory

https://drive.google.com/open?id=0ByoFGh573uhzWGx0aXNNNkY5Qms

• Detect objects in a given image using original YOLOv2

python predict.py -p ./test_imgs/multiple_stop.jpg -w ./weights/yolo-coco.weights -o ./

TODO List:

• Generate anchors using K-mean clustering on training data
• Convert DarkNet19 weights to Keras weights
• YOLOv2 Loss Function
• Hierarchical Tree
• Train on any custom data set
• Use MobileNet/DenseNet as feature extractor.

Acknowledgement

Thank you Dr. Christoph Merzt, Jina Wang, fellow scholars of RISS 2017 and Carnegie Mellon University for providing extraordinary support, resources and mentorship to help me complete this project.