<![CDATA[Surabaya Zoo is one of the conservation institutions in Surabaya, which has Sumatran elephants as a collection of endemic Indonesian animals. The Indonesian government protects this animal because of its endangered status. Having CCTV cameras installed in captivity helped us to create Elephas-SAM by utilizing Segment Anything Model (SAM) technology as the initial foundation for developing a system for monitoring animals in captivity with artificial intelligence (AI). Our investigations differ from past research in that we utilize 60 exclusive images obtained from CCTV footage in an elephant enclosure at Surabaya Zoo over a 30-day period instead of using publicly available datasets. The image set was partitioned into 30 instances taken under low-light settings (01:00 WIB) and 30 instances taken under high-light conditions (15:00 WIB). We perform the evaluation of SAM's prediction scores using the SAM-Point Prompt and SAM-Box Prompt techniques. It was found that, on average, the segmentation prediction scores for 30 low-light images are higher when the SAM-Point prompt is used (0.941) instead of the SAM-Box prompt (0.939), which is only a 0.002 difference. For a set of 30 vivid images, the SAM-Point Prompt produces a higher average score (0.989) than the SAM-Box Prompt (0.968), indicating a difference of 0.021. The results emphasize the effectiveness of using a SAM-Point prompt instead of a SAM-Box question to accurately forecast segmentation scores for items of Sumatran elephants under different illumination situations.]]>
