Artificial Intelligence or Machine Learning more specifically has a lot of potential. But its obscure and hard to understand. How can we go beyond just marketing

We provide custom ML research and development services with our expertise in vision and embedded development.

Lot of larger corporations publish exciting advances in ML, which we can leverage and build real use cases and create businesses value.

Here are some fields we are passionate about and to give a stronger understanding of what is possible

Classification

CIFAR10

Machines now exceed human accuracy at classifying images in various tasks. This can be used to classify different kinds of images

  • Recyclable or Non-Recyclable
  • Healthy food or Unhealthy Food
  • Threat or Not-Threat
  • My Dog or Neighbors Dog
  • Skirt or Pant
  • Cereal or Muesli

total number of classes can be large set as in retail products.

We can also classify multiple labels for the same image. like a dress can have these from model output - red, cotton, men, shirt, formal.

Detection

Identifying different objects is termed as Object Detection. Multiple instances of the came class can be identified.

  • Identify different fruits in an image and create recipe for smoothie
  • Count the different objects in an image
  • Detect presence of a person

We can then use other algorithms like optical flow to track the objects in the image.

Segmentation

Segmentation is classifying each pixel of the image to belong to a particular class

There are two kinds

  • Semantic Segmentation
  • Instance Segmentation

Semantic just classifies each pixels into a category, like tree, floor, road, car. etc.

Instance segmentation considers which instance the classified pixels belong to, example if there are group of tables, it will identify and segment each table.

Pose Detection

Pose estimation models can now track human poses in real time.

Use cases are immense and opportunities are real.

Gesture Recognition

Similar to pose detection, we can do palm and finger detection as well. What is more exciting is use of a temporal networks we can even classify different actions over a time frame, like gestures.

Google Open Source Model

Action Classification

We can also use time as a factor and investigate recurrent networks to classify different actions.

Generative Networks

Zolando Research

Generative Networks have advanced significantly. Reliably better with outstanding quality as shown above.

  • Generate fake person images
  • Transfer dress from one model to another demography
  • Make a person look old or young

You are the domain expert, we are the technology enablers. Lets connect to implement the future.

hello@someshwara.com