B2B Engineering Facts

Computer vision is a extremely specific, but at the same time broad field when you start talking about bespoke solutions. We get a fair amount of inquiries about projects nowadays, so to make it a bit easier for both our customers and us here is a short summary of what we have done or can do.

We currently are fully booked until 2016 Q2, so in case your project needs to be delivered before the end of Q2 we might not be able to fulfill your requirements.

When in doubt, feel free to contact us!

Main Products

Large Scale Image Retrieval/Search

This is our flagship product. We have a ready to deliver product for customer deployment, which we provide as a API for customers. Here is a list of things we will need from you:

  1. A file dump of the images you want to search.
  2. Metadata for said images. (e.g. if this is for a online store, we will need a mapping from the image file to your unique product code.)
  3. Estimation of how many concurrent users you expect.
  4. Whether search will happen via unique file uploads (e.g. Smartphone camera), readily available online images (e.g. Image URLs), or both.

There are products which image search are not good at, for example here is a picture of an iPhone 6S Plus, 6S, 6 Plus, and a 6. (Source: Macworld)

iPhone 6S Plus, iPhone 6, iPhone 6 Plus

The only visual cue we can work with is the [S] mark on the 6S and 6S Plus, which in a large scale search application is most likely going to get neglected. Here is another example: (Source: Know Your Mobile]

Galaxy S5 and S4

While we are not entirely sure, comparing to a test device we have we think the right one is the S4. For a image search engine with no metadata to work with, if the search results are correct that is because of what is on the screen, and not what the phone looks like. You get the idea. If there is not enough visual information to work with, it's most likely the search results are not going to be good.

However if there are enough visual cues to differentiate one from another, it's most likely to work. Get in touch with us (and don't forget to send sample images and queries, it really helps us determine if the project is feasible or not.) and we can discuss what we can do for you.

As for how many images we can have in our search engine, we can provide exact numbers with a NDA, but unless you are Amazon, Ebay, Flickr, or another huge service provider which has so many images it will be faster to Fedex us a huge box full of drives, our engine can host all of your images.

Style Search

An example application of this technology is currently used by one of our customers. You can download the app from the Google Play store, here and try yourself.

Imagine if you was in a store, and saw a flannel shirt on display. Unfortunately you are a man, and you happened to see this shirt in the women's department. You would like to know if said design is available in any of the men's brands. This technology lets you look for patterns of clothing in different categories and departments.

Style search (Source: Viva100)

Consulting Projects

Lens Correction

We have delivered a distortion correction engine to a customer for embedded endoscope applications. The core technology is the same for all lenses, so even if your application is not for endoscopes we may still be able to meet your needs with the same technology.

Deep Learning

We have done several projects with convolutional neural networks for various customers with various needs related to image classification and analysis for medical and biological applications.

Video Search

We have done video search projects using our image search technology as pilot research projects for customers.

What we do not do

Academic Collaboration

We are currently working with two universities for our academic research needs, and have no immediate plans to expand here for the time being.

Facial Recognition

When people here computer vision, this is one of the first things they associate it with. Unfortunately, we do not provide technology for facial recognition.

For basic applications, there is plenty of academic sample code out there which can be used.

For advanced applications which require actually determining who is in a given image, we simply have not had the time or resources to develop the technology. There are other vendors out there who provide such software.

Augmented Reality

We do not do augmented reality consulting projects. There are many vendors and libraries out there which you can readily consult for this.

Optical Character Recognition

While we have developed technology for applied optical character recognition (e.g. OCR for distorted/noisy images) we do not have core OCR technology in our portfolio.

Surveillance

We do not do projects involving mass surveillance of citizens, which may have privacy or human rights concerns.

Weapons

We do not do projects which involve weapons of any kind, including weapons of mass destruction.