Based on Object Detection for Multiple Object Detection
This project implements an object detection framework based on the YOLO V5s architecture recognize for multiple classes.
A total of 14 categories were annotated and used for model training, enabling accurate and efficient detection in subsequent inference tasks.
Under the configured training, the proposed model achieved approximately 90 % accuracy on the test dataset.
All images were manually annotated using bounding boxes corresponding to 14 predefined object.
The annotations were formatted according to the YOLO standard:
<class_id> <x_center> <y_center> <width> <height>
All coordinate values were normalized relative to image dimensions.
1.2 Dataset
The dataset was divided into training and validation subsets as follows:
Subset
Number of Images
Training Set
100
Validation Set
20
This partition was used to monitor generalization performance during training.
2. Model Architecture
The YOLO V5s model was adopted due to its balance between detection accuracy and computational efficiency.
This configuration is suitable for real-time application.
3. Training Configuration
The model was trained using the following hyperparameters:
Parameter
Value
Batch Size
16
Epochs
150
Image Size
640 × 640
These settings were selected to ensure sufficient training iterations for convergence while maintaining stable memory usage.
4. Hardware and Software Environment
Component
Version
PyTorch
10.2
GPU Acceleration for CUDA
10.2
The training process was performed using GPU acceleration to improve computational efficiency.
5. Performance
After training, the YOLO V5s model achieved approximately:
90 % accuracy on the test dataset
This result demonstrates that the trained model is capable of reliable object detection under the given dataset conditions.
This study demonstrates that a properly configured YOLO V5s model, trained on a labeled dataset of 14 object classes, can achieve high detection performance with limited training data.
The lightweight architecture makes the model suitable for deployment in practical real-world scenarios where computational resources are constrained.
