How Suning’s Facial Recognition Powers Unmanned Stores and Beats Global Benchmarks

From April 20‑22, 2018, QCon, the global software development conference organized by InfoQ, was held at the Beijing International Convention Center. Suning’s Silicon Valley Research Institute senior architect Wang Zhiguan presented the latest achievements of Suning’s facial‑recognition technology and its applications in unmanned stores and other scenarios.

Facial‑recognition technology identifies or verifies individuals from images or video. A complete system includes image acquisition, face detection, alignment, feature representation, and classification. Wang explained that deep learning has dramatically improved accuracy, and that Suning discovered the ResNet deep convolutional network can boost speed and performance. By integrating GoogleNet’s Inception module, which captures multi‑scale features, with ResNet, a new architecture further enhances model precision.

Rich application scenarios are essential for technology research. The industry believes algorithmic gaps are narrowing, so the focus shifts to deploying technology in specific industries and monetizing large‑scale data. Suning, with abundant data and scenes, has successfully applied facial‑recognition in unmanned stores, employee face‑entry to campuses, and store video recognition.

Unmanned stores: facial‑recognition enables seamless checkout, with payment times as fast as six seconds in Suning’s Store 2.0, now being rapidly replicated across multiple locations.

Employee face‑entry: facial‑recognition at gate cameras verifies identity in less than one second, offering a convenient alternative to card swipes with an error rate of only three in a million.

Store video recognition: the system captures store PV/UV, calculates foot traffic, sales conversion, and generates heatmaps of customer movement and product interaction.

Wang, a PhD from City University of Hong Kong and expert in facial‑recognition and computer vision, leads a team of engineers and researchers from AI, big data, and deep‑learning backgrounds. In March, Suning’s models ranked first and third respectively on the international LFW and MegaFace facial‑recognition benchmarks, achieving over 96% accuracy on MegaFace and 99.83% on LFW using improved ResNet and Inception‑ResNet architectures with multiple loss functions and fine‑tuning.

As a leader in smart retail, Suning continuously invests in computer‑vision technologies, believing that accumulated expertise, strategic decisions, and a suitable industry ecosystem will drive its vision technology to become a core innovation force in China’s smart‑retail transformation.