BETA
This is a BETA experience. You may opt-out by clicking here
Edit Story

Deep Learning-Enabled Image Recognition For Faster Insights

NVIDIA

More than two billion images are shared daily in social networks alone. Research shows that it would take a person ten years to look at all the photos shared on Snapchat in the last hour! That’s quite a tedious task and well past human capabilities.

Media buyers and providers experience difficulty organizing relevant content in groups, parsing components of images/videos, and defining the return on investment from generated content in an efficient way. Getting insights (i.e. metadata, composition, categories, color, etc) out of rich media — fast, accurately, and automatically — has long been a challenge.

NVIDIA has many customers and ecosystem partners tackling that problem, using NVIDIA DGX as their preferred platform for deep learning (DL) powered image recognition. One of the notable names among the ecosystem is Imagga, a pioneer in offering a deep learning powered image recognition and image processing solution, built on NVIDIA DGX Station, the world’s first personal AI supercomputer. Before we explore Imagga helping their customer PlantSnap further, let’s consider some of the key terms that are relevant to the image recognition domain.

Important Terminology Relevant to Image Recognition

  • Machine learning (ML) is a manual approach to achieve artificial intelligence. It allows us to approximate intelligence with algorithms and is a buzzword that we hear daily. ML uses algorithms to parse data, learn from it, and help predict outcomes.
  • Deep learning, a subset of of ML or a technique to implement ML, is quickly taking over industries as developers use the power of neural networks to discover insights. Automatically detecting patterns without human hand-coding features is the new holy grail.  
  • Computer vision enables computers to identify images, by using sensors and image processors to match human eyes’ capabilities. There is a slight overlap between ML and CV (see diagram above). CV has become important as it helps process the astonishing amount of visual imagery created every day.
  • Image processing is a method to perform some operations (enhancement or compression) on an image to extract useful information. ML can be used in both computer vision and image processing.
  • Image recognition is another term to articulate the process of identifying and detecting an object or a feature in images or videos. There are those who consider computer vision the same as image recognition, but computer vision is much broader and includes object recognition, character recognition, and text/sentiment analysis.  It’s common to associate image recognition with facial detection, but there’s much more to it. For example, here are some of the major features that Imagga offers:

  1. Classify/categorize scenes - automatically categorize photos; for example, sorting our personal photos as mountain vs. beach vs. pet etc. However, one of the more powerful capabilities of Imagga is to train a specific classifier for a specific use case and vertical, ranging from personal photos to plant identification to sorting garbage.
  2. Recognize the main object - identify the main object in an image.
  3. Auto-tag photos - label many objects with different keywords to help sorting photos and/or extracting statistics.
  4. Extract colors - identify representative colors in images, including separate recognition of foreground and background colors where needed.
  5. Analyze composition - automatically detect the most visually interesting areas in photos and eventually enable smart cropping of these areas of interest.
  6. Facial recognition faces - detect faces and cluster them in virtual persons, that can be used for better organization of personal photos and social media monitoring.
  7. Visual similarity search - extract signatures of the images in a collection that enables search among them based on visual and/or semantic similarity later on, as well as similar photos/products suggests.

Imagga customizes its solutions based on customer data and was one of the first to provide image auto-tagging as a service back in 2012, and then as part of its self-service cloud offering in 2014. They now serve more than 250 customers and 15,000 developers in more than 80 countries.

One of Imagga’s customers is Telluride, Colo., startup PlantSnap, which uses Imagga’s custom API to categorize plant images. With NVIDIA DGX Station, the purpose-built AI workstation, Imagga was able to extend the scope of the classifier behind PlantSnap for 120,000 species trained to 320,000. And that translates growing from 40 million to 90 million plant images collected worldwide for ML training. They sped the training service delivery by up to 88%, while maintaining the same level of accuracy.

Imagga has helped their customers recommend content, profile and analyze customers, and organize photos. To learn more about how Imagga uses DGX Station for the purpose of high-speed ML training, join us for an exclusive webinar with Georgi Kadrev, Imagga CEO and co-founder, on September 18, 2018, at 9am PST.

Resources: