Furniture is a household necessity that serves to complement a room. The design models of furniture are varied, ranging from general or mass-market designs to those made for specific needs. Sofas, tables, chairs, swivel chairs, TV cabinets, wardrobes, beds, and bookshelves are the items analyzed in this study. The fact that there are always more and better furniture models available is one of the reasons the researcher chose these furniture items. The goal of furniture object classification is to help categorize objects automatically, making it easier for users to quickly search for the furniture products they want, and the system can recommend items based on the classification results. This study uses 3 architectures with 3 testing scenarios. The architectures used are MobileNetV1, ResNet-50, and VGGNet-19. Scenario S1 uses image dimensions of 128x128 with 50 epochs, scenario S2 uses image dimensions of 128x128 with 100 epochs, and scenario S3 uses image dimensions of 128x128 + grayscale with 50 epochs. The accuracy results are differentiated according to the scenarios used, namely S1, S2, and S3. The detailed accuracy results are as follows: Scenario S1 using the MobileNetV1 architecture achieved an accuracy of 94.31%. The highest accuracy in scenario S2 was also achieved with the MobileNetV1 architecture at 94.31%. For the highest accuracy in scenario S3, MobileNetV1 achieved an accuracy of 72.61%. The fastest computation time for S1 was 871.97 seconds, for S2 it was 1763.04 seconds, and for S3 it was 436.32 seconds. Among the three architectures used, MobileNetV1 stands out as the best architecture in terms of accuracy and classification speed in this study