ViSenze takes visual search and image recognition to the next level

Guangda: "The better the GPU, the more advantage we can gain to accelerate our R&D"
Li: “The better the GPU, the more advantage we can gain to accelerate our R&D”

Every day, around three billion images and videos are uploaded online, creating a massive need to make them discoverable and searchable.

Backed by Rakuten and WI Harper, ViSenze took on this challenge to develop a solution that can help make sense of the ever-growing number of images.

The ViSenze team of web specialists and computer scientists with deep machine learning and computer vision experience, have since developed intelligent image recognition solutions for retailers and publishers.

Web becoming more visual
“We do visual search and image recognition for wider impact because the web is becoming more visual. Our solution provides the technology to understand pixel level content and then index them for easy discovery by people,” said Li Guangda, CTO and Co-founder of ViSenze, which has offices in the US, the UK, India, China, and Singapore.

The company started as part of NExT, a leading research centre jointly established between National University of Singapore and Tsinghua University of China.

Retailers such as ASOS, Rakuten and Uniqlo use ViSenze to convert images into immediate product search opportunities, improving conversion rates. Media companies use ViSenze to turn any image or video into an engagement opportunity, driving more new and incremental business.

GPU powers R&D
ViSenze uses the NVIDIA GPU to develop and power its solution.

“We chose the GPU as our technology because a few years ago, GPUs started to be used as a tool to accelerate deep learning research and development (R&D),” said Li.

“This is a great asset to improve our R&D efficiency. The better the GPU, the more advantage we can gain to accelerate our R&D,” he added.

The NVIDIA GPUs are used extensively to train models.

“We use convolutional neural networks to improve our feature interaction and detection performance. For our offline training and online inference, we also use GeForce cards to accelerate the processing of the image,” said Li.