这篇文章来自 nvidia.com。原始 url 是: https://blogs.nvidia.com/blog/2018/07/12/blazing-fast-image-identification/
以下内容由机器翻译生成。如果您觉得可读性不好, 请阅读原文或 点击这里.
Whether you’re shopping on Amazon, Etsy or any favorite digital boutique, seeing photos makes all the difference. But processing all these images is no small feat.
To ease the process, Imagga, a Sofia, Bulgaria-based startup and member of the nvidia 初始计划, provides image recognition and tagging services for companies. The company’s API can help image-intensive businesses improve photo search and delivery for customers.
Imagga’s technology can be applied to online retail to automate digital asset management and help boost advertising campaigns as well.
Identifying Mother Nature
Telluride, Colo., startup PlantSnap is using Imagga’s technology. PlantSnap’s app — available on Android and iOS — is used to identify plants from photos. PlantSnap users can take photos and the app will return information about the species within seconds, thanks to artificial intelligence.
To accurately identify flowers, seeds, trees, mushrooms and cacti from all over the world, PlantSnap’s algorithms needed to be trained using a vast quantity of data.
PlantSnap turned to Imagga to assist with the challenge of categorizing more than 320,000 different species of plants worldwide. For PlantSnap, it was crucial that its system was accurate and scalable. To meet PlantSnap’s needs, Imagga built a custom categorization API, which allowed the company to train its models 10x faster than previously possible.
Massive Training Data
To train the API, Imagga used over 90 million images, enabling it to maintain extremely high levels of accuracy, even when new plant species are introduced to the dataset. Imagga trained its networks using the NVIDIA DGX 站, which sped up its training time by 88 percent.
“Having a DGX Station to perform extremely large-scale training in a fraction of the time that we would need with other solutions is a huge advantage for us,” said Georgi Kadrev, co-founder and CEO of Imagga. “By training our networks on the NVIDIA DGX Station, the training time of our generic classifiers decreased from three weeks to just four days.”
Imagga is one of more than 2,800 startups in the NVIDIA Inception program. The virtual accelerator program provides startups with access to technology, expertise and marketing support.
Network with innovative startups in the program at GTC Europe, in Munich, October 9-11.