Gita, Kartika Diva Asmara and Kolbi, Zakia (2025) Implementasi Segmentasi Pakaian Menggunakan Metode Mask R-CNN. Project Report. [s.n.], [s.l.]. (Unpublished)
![[thumbnail of 5025211039_5025211049-Project_Report.pdf]](http://repository.its.ac.id/style/images/fileicons/text.png) |
Text
5025211039_5025211049-Project_Report.pdf - Accepted Version Restricted to Repository staff only Download (3MB) | Request a copy |
Abstract
Implementasi Mask R-CNN merupakan pendekatan untuk segmentasi objek fashion yang menggunakan dataset berisi 15.000 gambar. Mask R-CNN, dengan backbone ResNet101 dan arsitektur Feature Pyramid Network (FPN), digunakan untuk mendeteksi dan menghasilkan mask segmentasi dengan presisi tinggi. Penelitian ini bertujuan untuk mengevaluasi kinerja model dalam mengenali dan memisahkan berbagai jenis pakaian berdasarkan citra. Hasil pengujian menunjukkan bahwa konfigurasi model dengan jumlah epoch dan langkah per epoch yang terbatas belum mampu mencapai kinerja optimal yang hanya mampu mencapai nilai rerata IoU dan Accuracy hanya sebesar 36.93% dan 90.75%. Sedangkan rata-rata IoU dan Accuracy untuk metode superpixel sebanyak 49.29% dan 92.12%. Faktor seperti jumlah epoch, kualitas dataset, dan pengaturan hyperparameter berperan signifikan dalam mempengaruhi hasil segmentasi. Penyesuaian lebih lanjut diperlukan untuk meningkatkan akurasi dan efisiensi model. Penelitian ini memberikan kontribusi pada pengembangan teknologi segmentasi objek di industri fashion dan e-commerce, khususnya dalam analisis visual berbasis citra.
============================================================================================================================
Mask R-CNN implementation is an approach for fashion object segmentation that using a dataset of 15,000 images. Mask R-CNN, with the ResNet101 backbone and the Feature Pyramid Network (FPN) architecture, is used to detect and generate a high-precision segmentation mask with high precision. This study aims to evaluate the performance of model in recognizing and separating different types of clothing based on images. The results of The test results show that the model configuration with a limited number of epochs and steps per epoch has not been able to achieve optimal performance. model configuration with a limited number of epochs and steps per epoch has not been able to achieve optimal performance which is only able to achieve an average value of IoU and Accuracy are only 36.93% and 90.75%. While the average IoU and Accuracy for the superpixel method are 49.29% and 92.12%. Factors such as number of epochs, quality of quality of the dataset, and hyperparameter settings play a significant role in influencing the segmentation results. segmentation results. Further adjustments are needed to improve the accuracy and efficiency of the model. model. This research contributes to the development of object segmentation technology in the fashion and e-commerce industries, especially in image-based visual analysis. fashion and e-commerce industries, especially in image-based visual analysis.
Actions (login required)
 |
View Item |