YoloKeras-Detection

This project was developed by António Pedro Silva Matos for object detection using the YOLO (You Only Look Once) model implemented with Keras. The goal is to provide a unified platform to explore, test, and compare the performance of this model across diverse datasets and scenarios.

Summary

1. Introduction
2. Overview
3. Functionalities
4. Configuration
5. Requirements
6. Configuration
7. Preperation
   • Download
   • Installation
8. Contribution
9. License

Introduction

This repository presents an innovative platform for object detection, using the YOLO model implemented with Keras. Developed to provide a detailed comparison and accessible implementation, the project aims to help researchers and developers explore and evaluate the effectiveness of the model in diverse scenarios.

Overview

Object detection is a crucial area of ​​computer vision, with applications in surveillance, autonomous driving, video analysis, and more. In this repository you will find:

• Implementation of YOLO using the Keras framework.

• Detailed performance comparisons, including metrics such as precision, recall, and inference time.

• Guides to help understand and replicate results.

Functionalities

• YOLO (You Only Look Once): An object detection model known for its speed and accuracy.

• Performance Comparison: Scripts to evaluate and compare the performance of models in terms of different metrics.

• Datasets: Support for multiple popular datasets such as COCO, Pascal, VOC, and others.

Requirements

• Python 3.7 or higher

• Keras 2.x

• TensorFlow 2.x as Keras backend

• Additional libraries listed in requirements.txt

Configuration

This project uses Yolov4, that includes the files yolov4.cfg, yolov4.weights and coco.names

Before execute the code, verify if the files are present on directory after downloaded the repository

• Yolov4.cfg: Model configuration
• yolov4.weights: Model weights
• coco.names: File with the names of the objects that the model was trained to detect

Preperation

Download

Before downloading the repository, we need to download the model settings and weights so that we can use the program itself. I'm using the normal YoloV4, but if your PC doens't have to much memory. I recomend to use tiny YOLO, where will the links to download be, however in the code they will have to modify the file name so that it can run correctly.

• YoloV4 weights
• YoloV4 config
• YoloV4 classes

For tiny Yolo:

• YoloV4-tiny config
• YoloV4-tiny weights

Installation

1. Clone repository:

      git clone https://github.com/AntonioPedro07/YoloKeras-Detection.git
   

2. Navigate to the project directory:

      cd YoloKeras-Detection
   

3. Install dependencies:

      pip install -r requirements.txt
   

Contribution

Feel free to open issues and submit pull requests. Contributions are welcome!

We consider exploring the Darknet YOLO ecosystem for valuable insights and testing different dataset versions and models. Our goal is to improve object detection using the official Darknet repository. We plan to create a personalized dataset with relevant objects to improve the mobility of the visually impaired, aiming to significantly increase detection accuracy.

For this code we utilize a flexx2 depth camera from Pmdtec

License

YoloKeras-Detection itself is released under the MIT License (refer to the LICENSE file for details). Portions of the code are borrowed and given by pmdtechnologies and from their RoyaleSDK