Seq_nms_YOLO

Membres: Yunyun SUN, Yutong YAN, Sixiang XU, Heng ZHANG,

---

Introduction

This project combines YOLOv2(reference) and seq-nms(reference) to realise real time video detection. [Tutorial develop for Ubuntu].

The lines attached under this colour in this repository are for running under the Ubuntu terminal.

Steps

1. Open a terminal.

2. Create a virtual environment with python 2.7:

   • conda create --name EnvExample python=2.7.
   • conda activate EnvExample.

3. Clone the github repository in the folder you would like to have the project:

   • git clone https://github.com/carlosjimenezmwb/seq_nms_yolo.git.

4. Go inside the project:

   • cd seq_nms_yolo.

5. Make the proyect using the command:

   • make.

6. Download the yolo.weights and tiny-yolo.weights:

   • wget https://pjreddie.com/media/files/yolo.weights.
   • wget https://pjreddie.com/media/files/yolov2-tiny.weights.

7. You must have the following libraries installed (with indicated versions)

   • cv2 pip install opencv-python==4.2.0.32.
   • matplotlib pip install matplotlib.
   • scipy pip install scipy.
   • pillow conda install -c anaconda pillow.
   • tensorflow conda install tensorflow=1.15.
   • tf_object_detection conda install -c conda-forge tf_object_detection.

8. Copy a video file that you want to use to the video (/seq_nms_yolo/video) folder, for example, 'input.mp4';

9. Go to the directory /seq_nms_yolo/video and run video2img.py and get_pkllist.py:

   • python video2img.py -i input.mp4.
   • python get_pkllist.py.

10. Export the paths:

    • export PATH=/usr/local/cuda-10.1/bin${PATH:+:${PATH}}.
    • export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/usr/local/cuda-10.1/lib64s.
    • export LIBRARY_PATH=$LIBRARY_PATH:/usr/local/cuda-10.1/lib64.

11.Return to root folder and run yolo_seqnms.py to generate output images in video/output: * python yolo_seqnms.py.

12. If you want to reconstruct a video from these output images, you can go to the video folder and run img2video.py:
    • python img2video.py -i output.

And you will see detection results in video/output

Reference

This project copies lots of code from darknet , Seq-NMS and models.