Multi-Cloud
Smart Vision AI
High-Precision Furniture Recognition & Classification Vision Solution Optimized for Real-World Environments
This project is a case study of building a YOLOv8-based object detection model to recognize and precisely classify furniture objects in real time within retail and residential environments.
TOSKY directly built a high-quality image dataset of 689 images covering major furniture categories including chairs, desks, and sofas reflecting IKEA's design characteristics, establishing the foundation for model training.
Beyond simple training, an optimization process was conducted to address dataset imbalance by precisely restructuring the train/validation/test ratio to 72:14:14. As a result, GPU-based 100-epoch training achieved dominant performance of 97.8% for chairs and 94.7% for sofas across individual classes, securing an overall mAP of approximately 95.4% — a level of accuracy suitable for real-world deployment.
The model was further refined to overcome the initial data limitation of static image bias and operate stably across various angles and background conditions, completing the technical foundation for instantly identifying furniture information in actual store promotional videos or live commerce environments.
Key Features
01
YOLOv8-Based High-Precision Object Detection
Leverages state-of-the-art deep learning architecture to analyze furniture shapes and features, classifying types and identifying locations in real time
02
Verified mAP 95.4% Recognition Performance
Through data optimization and iterative training, achieved dominant accuracy above 90% across all classes including chairs, desks, and sofas
03
Real-Environment Adaptive Model Refinement
Completely resolved data bias to ensure stable operation in actual store environments with numerous variables such as background noise and camera angle changes