YOLOv2 for real-time video Object Detection

This is modified from the PyTorch implementation of YOLOv2(by Long Chen https://github.com/longcw/yolo2-pytorch).
For the video process, I used OpenCV and FFmpeg for real-time and frame-based transformation.
Real-time Object Detection works on 720p videos of different types.

The model is mainly based on darkflow and darknet.

Used a Cython extension for postprocessing and multiprocessing.Pool for image preprocessing. Testing an image in VOC2007 costs about 13~20ms.

For details about YOLO and YOLOv2 please refer to their project page and the paper: YOLO9000: Better, Faster, Stronger by Joseph Redmon and Ali Farhadi.

Requirements

• python 3.6
• Anaconda3
• pytorch 0.3.0+
• gcc
• cuda 8.0+    

Installation and apply Object Detection on Videos

1. Clone this repository

   git clone git@github.com:judichunt/yolo2-pytorch-realtime-video

2. Build the reorg layer (tf.extract_image_patches)

   cd yolo2-pytorch
   ./make.sh

3. Install opencv

   conda install -c conda-forge opencv 

4. Download the trained model yolo-voc.weights.h5
   Set the model path h5_fname, and input_dir filename of video file(.avi, .mpeg, ...) in realtime_OD.py

5. Run python realtime_OD.py. The real-time Object Dectection video will be played as running the model.

   And the output video will be saved as the same type as your input video.

Demo video

Training YOLOv2, Evaluation, Training on your own data

Follow the instructions in
YOLOv2 in PyTorch.