Skip to content

Latest commit

 

History

History
executable file
·
90 lines (60 loc) · 2.92 KB

README.md

File metadata and controls

executable file
·
90 lines (60 loc) · 2.92 KB
Raw

Real-Time Video Recognition AI with auto-webscraping trainer + alerts, YoloV3

By: Nasrudin Salim

Real-time object detection and classification with YoloV3

Test:

What is this:

  • Automatically trains a model via webscraping image search results on a video recognition classifier with transfer learning.
  • Enter a label, then enter a list of search queries to google for. It will then google for those search terms and fine-tunes a pretrained classifier.
  • Detect objects as well as output alerts/colors differently if an object in your "alert list" is found.
  • Can be performed on a video stream in real-time.
  • Can be performed on a live-camera stream in real-time.

Prerequisities

  1. Anaconda 3
  2. Python 3
  3. CUDNN and CUDA ToolKit Installed
  4. GPU with CUDA 9.0 and above support

Installation Instructions:

  • A conda environment file has been provided, please make use of environment file to install the necessary requirement packages. conda create -n "YourEnvironmentName" -f environment.yml
  • Project uses Cython, so build that too. python3 setup.py build_ext --inplace

Remember to download pretrained weights if you need to See training section

Usage

Getting Data

On the webscraper, indicate the labels in labels.txt, and change the parameters in config.py if needed then in terminal, type

	python3 downloadimages.py

Training

Once images are downloaded. You can download pretrained weights here: darknet Or you can continue training your weights if you've done this before Edit the parameters in train.py and then in batch type:

	python3 train.py -g

-g : Whether to use GPU
-e #: Epochs (optional)
-s #: Save Rate (optional)
-e #: Epochs (optional)
-lr #: Learning Rate (optional)
-b #: Batch (optional)

Testing/Using

Usage:

python3 run.py "path_to_video_file" -g -s "test.avi"

Explanation

Whether to use a video file, supply path
If not video file, assume to use camera feed.
-g /--gpu : Optional, whether to use GPU ( Defaults to config)
-s /--save: Optional, whether to save results to a video file and where (Defaults to config)

Configs

There is a config.py provided which helps to set the defaults for the following:

  1. Edit the path to weight file
  2. USE GPU? Bool
  3. SAVE Result video file? Bool

You will be asked for the path of the video file. You can adjust the parameters as well as the paths of the weights by opening up the py files.

Setting Alerts

You can set alerts by editing the text file "alerts.txt" when a label found in this text file appears, it will generate an alert by drawing the box red and displaying "Alert x found in footage" when testing.

HELP AND CONTRIBUTION

This project needs your help and contribution
The Training Script needs work.
The Alerts and Pinging requires work to finish.
Downloading of images from google and seamless Training