Evidence record · UDINUS · SDGs
- Unit
- Rektorat UDINUS
- Period
- 2024
- Goals
- Record
- b87c0b2d-0f78-491c-b5c9-083e4039ebfe
- Published
- 2026-10-30
Universitas Dian Nuswantoro conducted research on the effect of the LAB color space and NASNetMobile fine-tuning on model performance for crowd detection. The activity involved UDINUS researchers Rafid, Muhammad; Luthfiarta, Ardytha; Naufal, Muhammad; Fahreza, Muhammad Daffa Al; and Indrawan, Michael, who participated in the research. The study was situated in the context of the COVID-19 pandemic, in which computer vision plays a crucial role in crowd detection and supports crowd restriction policies intended to mitigate virus spread.
The research analyzed the use of the RGB LAB color space in advance to improve NASNetMobile performance. The fine-tuning process involved freezing layers in various NASNetMobile base model variations. The results showed that the model using the LAB color space outperformed the model using the RGB color space. The average accuracy was 94.68% for the LAB color space, compared with 94.15% for the RGB color space.
Across the test iterations, the highest performance for the NASNetMobile model occurred when 10% of the layers from the back were frozen for both the LAB and RGB color spaces. Under this configuration, the LAB color space achieved an accuracy of 95.4%, while the RGB color space achieved an accuracy of 95.1%. This activity contributes to the stated goal through research addressing computer vision for crowd detection and its relevance to crowd restriction policies. UDINUS presents the activity as research on model performance, color-space use, and fine-tuning for crowd detection.