Scylla AI Sets New Computer Vision Benchmark as SCYLLA-MOT Ranks #1 on Global MOT17 Tracking Challenge

AUSTIN, TX — September 1, 2026 — Scylla AI, a global leader in real-time physical security and threat detection, has recently announced that its latest computer vision model, SCYLLA-MOT, has achieved the #1 ranking on the international MOT17 Multi-Object Tracking benchmark.

This milestone follows Scylla’s 2025 breakthrough, where its ScyllaNet architecture earned the top position on the COCO Detection Challenge. Together, these consecutive achievements establish Scylla AI as an industry leader across both core disciplines of computer vision: high-precision object detection ("what is in the frame") and continuous temporal tracking ("where did it go and is it still the same object").

Redefining Precision in Complex Video Environments

Evaluating tracking algorithms across crowded scenes, severe occlusions, and shifting camera perspectives, the MOT17 benchmark tests a model's ability to maintain identity continuity over time. SCYLLA-MOT set a new global benchmark by recording:

● 69.17 HOTA (Higher Order Tracking Accuracy) ● 81.83 MOTA (Multiple Object Tracking Accuracy) ● 87.35 IDF1 (Identity Preservation) ● 708 Identity Switches (a 39% reduction compared to fourth-ranked NVIDIA SPAM)

In physical security contexts, minimizing identity switches prevents fragmented tracks, ensuring continuous, reliable situational awareness when following individuals through high-density environments.

"At Scylla, continuous improvement of our AI models is not optional. It is one of our fundamental engineering priorities," said Albert Stepanyan, CEO and Founder of Scylla. "We operate in security, where the accuracy, consistency, and performance of a model can ultimately contribute to decisions involving people’s safety. Reaching #1 on MOT17, following our #1 result on COCO, provides independent, measurable evidence that this commitment is delivering real-world results.

Translating Research into Physical Security

By mastering both object detection and multi-object tracking, Scylla AI equips security operators with a richer layer of context, transforming raw video feeds into continuous, actionable intelligence.