{"id":190,"date":"2025-10-03T21:26:48","date_gmt":"2025-10-03T18:26:48","guid":{"rendered":"https:\/\/www.mkensari.com\/?page_id=190"},"modified":"2025-10-03T21:26:48","modified_gmt":"2025-10-03T18:26:48","slug":"sam2-rim-segmentation","status":"publish","type":"page","link":"https:\/\/www.mkensari.com\/index.php\/sam2-rim-segmentation\/","title":{"rendered":"SAM2-Powered Semi-Automated Data Generation for Vehicle Rim Segmentation and Custom Model Training"},"content":{"rendered":"
\n

SAM2-Assisted Rim Detection Process<\/h3>\n
\n \"SAM2-assisted<\/p>\n

SAM2-assisted wheel detection<\/em><\/p>\n<\/p><\/div>\n

\n \"Detection<\/p>\n

Detection of the metallic rim area using SAM2 and a custom model<\/em><\/p>\n<\/p><\/div>\n

\n \"Segmentation<\/p>\n

Segmentation result: selected metallic rim region<\/em><\/p>\n<\/p><\/div>\n

\n \"Before
\n \"After<\/p>\n

Segmentation result: replacing the rim with a provided PNG<\/em><\/p>\n<\/p><\/div>\n

Abstract<\/h2>\n

\n This study demonstrates how foundation segmentation models like SAM2 can be used to create high-quality datasets
\n for training specialized AI models. As a case study, we developed a model that segments only the rim region
\n (excluding tires) in vehicle images. After detecting and cropping vehicle wheels using SAM2, rim masks were
\n generated using classical computer vision algorithms (ROI-based analysis, Hough Circle detection, etc.).
\n Manual labeling was performed when automatic methods failed. The trained custom model achieved remarkable
\n success on real vehicle images. This approach can be generalized for many other specialized AI scenarios.\n <\/p>\n

1. Introduction<\/h2>\n

\n Specialized part segmentation is a critical need in fields such as virtual try-on, e-commerce, and automotive.
\n Traditional dataset creation is costly, but foundation models like SAM2 make the process semi-automatic.
\n This study presents an approach where general models are used as teachers, followed by training specialized
\n student models with high accuracy.\n <\/p>\n

2. Related Work<\/h2>\n

\n Segment Anything (SAM\/SAM2) models by Meta represent progress in zero-shot segmentation.
\n SAM2 produces improved mask quality and box-based predictions compared to the original SAM model.
\n Classical methods like Hough Circle were used here because they work well on cropped images.
\n To date, there has been no direct application of SAM2-supported custom model training.\n <\/p>\n

3. System Architecture and Methodology<\/h2>\n

3.1 Overall System Architecture<\/h3>\n