Comparison of VGG16, MobileNetV2, InceptionV3, ResNet50, and Custom CNN Architectures for Furniture Image Classification
Epiphany Shavna Gracia, Nurul Anisa Sri Winarsih · Infotekmesin · 2025
The rapid development of technology in the current digital era is driving increased demand across various sectors, including the furniture industry. Classifying furniture images is one of the critical challenges in image processing and computer vision, mainly due to the diversity of types. This research aims to understand how pre-trained models can affect image classification accuracy using furniture dataset results. This study uses five CNN architectures and focuses on comparing the performance of a custom architecture with four pre-trained architectures, namely VGG-16, MobileNetV2, InceptionV3, and ResNet-50, using furniture images that have five classes such as chairs, tables, cabinets, sofas, and mattresses. The research results show that the models produced by the pre-trained architectures provide higher accuracy and performance, with VGG-16 reaching 97%, MobileNetV2 at 96%, and InceptionV3 and ResNet-50 at 98%. Meanwhile, the custom model only achieved an accuracy of 85%. This research shows that using pre-trained model algorithms significantly improves performance in image classification.