FasterRCNN object detection

This application enables object detection in images and videos using Faster R-CNN models. It utilizes the PASCAL VOC dataset, allowing users to choose and download specific editions as needed. The application allows for dynamic data transformations, including rotation, resizing, and normalization. Users can perform training and testing on different backbones of Faster R-CNN models. They can also load and save custom models. Object detection on loaded images and videos is supported, providing confidence values. Evaluation of training effectiveness is facilitated through graphs displaying loss, accuracy, and the mAP metric (mean Average Precision). Results, including object detection outputs and training graphs, can be saved.

Screenshots

Main application view	Training results(*)

Object detection on image	Object detection on video

*These training results are provided for demonstration purposes only. They serve as examples to showcase the functionality of the application.

Getting Started

These instructions will get you a copy of the project up and running on your local machine.

Prerequisites

• Python 3.9 or later
• Pip
• NVIDIA GPU with CUDA support (optional): If you want to use the GPU version of the application, make sure your system has the appropriate driver and CUDA Toolkit installed compatible with your GPU version. Version 12.1 is required.

Installing

1. Clone the repository:

   git clone https://github.com/Dawid-Nowotny/FasterRCNN-object-detection.git

2. (Optional) Create a virtual environment (recommended):

   # Windows
   python -m venv venv
   
   # Linux/macOS
   python3 -m venv venv

   Activate the virtual environment

   # Windows
   venv\Scripts\activate
   
   # Linux/macOS
   source venv/bin/activate

3. Install the required dependencies using pip:

   pip install -r requirements.txt

4. Run the application:

    python main.py

   By default, the application interface is displayed in Polish. If you prefer to run the application in English, use the
   --en flag when launching:

    python main.py --en

   This will start the application with the interface in English.

Functionality

Dataset

• Select PASCAL VOC version and download it automatically
• Save the dataset to a cache file to speed up loading in subsequent application runs
• Display an example from the dataset
• Clear the loaded dataset
• Remove saved cache files

Model

• Load one of the Faster R-CNN models with specified backbone for training
• Load a model from a file
• Save a model
• Clear the loaded model

Training

• Set optimizer parameters
• Set scheduler parameters
• Set training parameters
• Train the model
• Display training results

Detection

• Recognize objects in a loaded image or video
• Replay the video
• Clear the currently displayed image or video
• Export image or video with bounding box annotations