BigData_Project: Blur faces and license plates using deep-learning

This project was worked with 서수진, 원동균, 이민지, 전여진

These codes were built on Ubuntu 18.04 enviroment(python_3.6, tensorflow_1.14.0, django_2.0.13, opencv-python_4.1.0.25 and ngrok_2.3.34) with GTX-1050ti

We trained face and license plate images with YOLO-v3

face: wider face(3226), license plate: AOLP(2049), MediaLAb LPR(590) and self collection(327))

Getting Started

Our weight was uploaded on here

You must move weight file to "./YOLOv3_TensorFlow/checkpoint" and modify restore_path argument of test_single_image.py at 43th row

1. Run the server in your project directory

$ python3 manage.py runserver

2. Activate server using ngrok on other terminal
   • Copy the address of 'Fowarding' row on terminal after execute below code (ex. d61b0f6fngrok.io)

$ ./ngrok http [port number] (maybe 8000)

3. Modify setting.py in FirstProject directory

   • Paste the address to ALLOWED_HOSTS at 28th row

4. Blur a image

   • Open the address on your browser
   • If you want blur the image, click icon
   • After click "파일선택" button, select the image to blur
   • Click convert button
   • If you want to download blured image, click icon
   • If you want to go home, click icon
   • We don't produce to blur the video on the web. If you want to blur the video, you must execute video_mosaik.py on your computer not our web
   • If any path(in code) is written by uni-code, it doesn't work because of open-cv

Result

If you input a image, the image is converted like below