Providing the brain behind electronic eyes
XjeraLabs relies on NVIDIA Tesla technology to deliver automated, accurate video analytics to drive business and smart city insights.
XjeraLabs aims to be the leading artificial intelligence (AI) video analytics company in the world. Established in 2013 in Singapore, the company won the Singapore IMDA Video Analytics Challenge 2015 for its outstanding work in the field. The company provides AI-based visual computing appliances (VCAs) for commercial use, and also has the expertise to develop custom high-performance video analytics solutions. XjeraLabs solutions cover quantification and classification of objects, attribute detection, face recognition, license plate recognition, tracking of persons and vehicles of interest, behavior and event detection.
XjeraLabs is collaborating with IMDA for various Smart Nation projects in Singapore for the government sector as well as with commercial firms such as a theme park and a chain of sushi restaurants in Singapore. One of its latest projects is to provide smart solutions via AI-based VCAs to Sentosa Development Corporation to solve daily operation bottlenecks.
XjeraLabs saw an opportunity to disrupt the field of video analytics, which traditionally relies on human monitoring or handcrafted computer vision algorithms, by applying artificial intelligence-based deep learning technology.
“Most companies rely on traditional algorithms, and accuracy suffers,” explained Grace Wei, VP, Project Management, XjeraLabs.
The company tried traditional CPU-based video analytics, but felt that the option is too slow and expensive when it came to achieving the levels of accuracy and stability that XjeraLabs’ customers required.
“The only way to make a product with high accuracy is to use deep learning algorithms. We can run the algorithms on CPUs but it would be very intensive computationally. We would need many units, and the price points rapidly become unacceptable to customers,” said Wei.
Wei added that stable technology was critical.
“Most of our customers are from the government, for whom all hardware must run 24×7. Production quality and reliability are very important,” she added.
XjeraLabs had used NVIDIA GPUs in the past, and found the NVIDIA Tesla P4 Inferencing Accelerator to be ideal for its requirements. The company is deploying the Tesla P40 Accelerator in conjunction with an optimized HPE platform as its main solution for deep learning, and then running deep-learning workloads with the Tesla P4 in its dedicated video analytics appliances.
Thanks to a collaboration between XJERA, NVIDIA and HPE, XjeraLabs’ deep learning training with the Tesla P4 requires approximately 11x to 15x fewer system resources than an academic or commercial deep learning solution. Tests also showed that using the Tesla P4 for processing is up to 150x faster than the CPU-only deployment that XjeraLabs had used previously.
“The NVIDIA GPU is the dominant hardware for deep learning. We decided on the NVIDIA Tesla P4 as it is well known as an innovative and cost-effective solution for advanced deep neural networks learning technology,” said Wei. “It is both stable and versatile.”
The Tesla P4 has ensured that XjeraLabs’ video analytics solutions meet customers’ requirements for accuracy, speed, pricing and stability. The company’s appliances can process data from multiple camera feeds in real-time, quantifying and classifying objects accurately while cutting down on the amount of post-installation fine-tuning that is typically required to ensure system robustness. The lower turnaround time has allowed XjeraLabs to implement custom solutions more quickly.
“The Tesla P4 is able to boost compute efficiency running deep-learning workloads, enabling smart, responsive AI-based services while keeping the system stable for heavy duty long-term deployment,” said Wei. “We’re more confident that our appliances are highly accurate. They also require minimal fine-tuning after installation now. With Sentosa we have achieved more than 95% accuracy across a range of different scenarios.”
XjeraLabs has been very pleased with the support from NVIDIA as well.
“No matter which deep learning framework we are using, even our own, NVIDIA’s team helps us optimize our algorithms for speed and accuracy, and advises us on what is technically feasible,” Wei said.