yolov4-Object-Detection-and-UI-Code-for-Deployment

This repo contains a wide range of custom functions & UI code for real-time detection & analysis of objects using YOLOv4. The UI can take a video (or CCTV streaming) as input, and will stream the output object detection in UI, along with time series count. Ready for deployment in the GPU servers for real-time custom object detection.

Techniques Used

Modeling Techniques: CNN object detection, Darknet53, Yolov4, Transfer Learning, Computer Vision.
Image & Video processing techniques: Opencv, GStreamer
Tech Stack: Python.
Libraries: Tensorflow 2.3.0, Scikit-Learn, Keras, matplotlib.
GUI techniques: Flask, HTML, JavaScript, RTSP.

A demo of the detection & analysis of objects on conveyor belt!

Currently Supported Custom Functions and Flags

Counting Objects (total objects and per class)
Print Info About Each Detection (class, confidence, bounding box coordinates)
Time-series plot for the detailed analysis (Creates the time-series plot on the object count in each frame)

Models

Here we have 3 models for 3 different use cases.

./checkpoints/yolov4-416: For pizza detection on conveyor belt. The original Darknet weights, trained on COCO dataset.
./checkpoints/yolov4-custom_tire_2000-416: For tire detection on conveyor belt. Trained on 1500 tire images, collected from Google Open Images dataset.
./checkpoints/yolov4-obj_cup_last-416: For broken cups detection from good ones. Downloaded 300 broken cups & good cups from Google images, labeled using labelbox, and then trained for 2 classes.

The original Darknet model weights are available online. Due to large size of the other model weights those are not uploaeded here, if you want the models weights or training datasets please let me know.

Getting Started

Folder structure

main_ui.py is the Flask app that launches the UI application and calls necessary backend functions
./templates/index.html is responsible for the UI
If the user provides an image as input detect_img_ui.py is called for object detection in the image, and if user provides a video as input detect_video_ui.py is called for object detection in the video.
./saved_detections/ consists of quite a few examples of yolov4 detections on pizza & tire on conveyor belt and broken cups.

└── Yolov4-Object-Detection-and-Custom-UI
       ├── checkpoints   <-- contains trained models
       |   ├── yolov4-416
       |   ├── yolov4-custom_tire_2000-416
       |   └── yolov4-obj_cup_last-416
       ├── core  < -- contains utility functions 
       |   └── utils.py
       ├── data   <-- contains images/videos for input to UI
       |   ├── classes
       |   ├── images
       |   └── videos
       ├── detections 
       ├── saved_detections <-- few examples
       ├── static
       ├── templates
       |   └── index.html   <-- responsible for UI
       ├── detect_img_ui.py
       ├── detect_video_ui.py
       ├── main_ui.py
       └── requirements.txt

Run the UI application

### Install the necessary dependencies from `requirements.txt`
### Launch the Flask app
python main_ui.py

GUI Navigation

The Flask app will be running from 5000 port, so please go to localhost:5000 in your browser
Import the weights and image/video: You can select yolov4-Pizza in the weights dropdown, choose pizza_radmaker1.mp4 for filename and press Load weights & file button, it will import the yolov4 weights and the pizza video in the backend.
Press start object detection button, the object detection will start in GUI in a few seconds.
It creates the time-series analysis plot for video input files, this feature can be switched on/off by setting is_time_count to True/False in main_ui.py.

A screenshot of the UI that detects defective items!

This project has been inspired by AIGuys, so I would like to thank him. If you have any question please feel free to connect me in Linkedin. Happy Learning!

saha0073/Yolov4-Object-Detection-and-UI-Code-for-Deployment