Skip to content

Latest commit

 

History

History
93 lines (81 loc) · 3.76 KB

README.md

File metadata and controls

93 lines (81 loc) · 3.76 KB
Raw

🐕 Stray Dogs Detection System 🐕

I. Introduction

Motivation

  • High population of stray dogs in Karachi leading to safety concerns and conflicts between humans and stray dogs.

Objective

  • Develop a stray dog detection model to identify and locate stray dogs across the city.

Solution

  • Create a system using AI and object detection to detect stray dogs and provide their location to authorities.

Benefits

  • Efficiently capture stray dogs, enabling authorities to sterilize and vaccinate them, reducing fear and promoting harmony between humans and stray dogs.

II. The Dataset

III. Model Training

  • Download the pre-processed dataset from the drive link

  • Upload it on Colab, or your own Drive.

  • If you uploaded it on Drive, then mount your Google Drive in your Colab file, else just upload the dataset's zip file on Colab (it might take some time).

  • Unzip the file using: 

     !unzip <path_to_dataset.zip_file>

  • Clone Yolov5's GitHub, cd to yolov5 and install the requirements using: 

    !git clone https://github.com/ultralytics/yolov5
%cd yolov5
%pip install -r requirements.txt

  • Upload yolo5s.pt and custom_dataset.yaml on your Colab

  • Train your own model using the command: 

    !python train.py --img 640 --cfg /content/yolov5/models/yolov5m.yaml --hyp /content/yolov5/data/hyps/hyp.scratch-med.yaml --batch 32 --epochs 50 --data /content/custom_dataset.yaml --weights /content/yolov5s.pt  --workers 24

    I trained for 100 epochs initially but 90 epochs gave the best result.

  • Save the trained model. You can find it at /contents/yolov5/runs/train/exp/weights/best.pt

  • The trained model is the best.pt file in this repo, but you can also find it at Drive-link

  • Apply object detection on a video/image file using:

    !python detect.py --img 640 --source <path_to_source_file> --conf 0.5 --weights /content/yolov5/runs/train/exp/weights/best.pt

Reminder: you can just use some other dataset, or change the parameters as per your preference to train your own custom object detection model :))

IV. Model Deployment Setup

  • This project might only work on your system if you have a GPU (else it will only work for image files and not the video ones)
  • You can replace the best.pt file with your own custom object detection model.
  • cmd: 
     cd <path_to_your_fav_directory>
    git clone https://github.com/tselane2110/stray-dogs-detection-system
    cd <path_to_the_cloned_repository>
  • create 4 folders in static\ :
    • images
    • predicted_images
    • videos
    • predicted_videos
  • create a virtual environment by executing the following command in your cmd 
    python -m venv virtual
  • activate the virtual environment by executing the following command in your cmd 
    virtual\Scripts\Activate.ps1
  • install the required libaries by executing the following command in your cmd 
    pip install -r requirements.txt
  • run the following command in your cmd 
    python main-app.py
  • go to 192.168.1.101:5000 on your browser
  • choose a file then upload it
  • that's it for now, have fun!

Note: This is an ongoing project, so Ill be updating it from time to time
Also, if you encounter any issue, kindly do let me know so I can fix it!