deep learning in the wild
Deep Learning in the Wild

LobsterReal-time stock photos

Describe what your project/business does. Which problem are you solving?

Lobster is an AI-powered platform which enables brands, agencies and the press to license visual content directly from social media and cloud archives, offering an alternative to ‘staged’ stock photography.

Can you describe your tech stack, what libraries hardware / architecture are you using and why?

For our machine learning tasks we use Python, C++, Tensorflow, word2vec and OpenCV, and for the training models we use our computing resources. In the near future we also plan on renting cloud resources.

What were some of the biggest challenges technical or business related you encountered and how did you overcome them?

The biggest challenge was to handle content without metadata. For example: Photos from Instagram include tags and names, so it’s easy to search for them, but photos from Dropbox don’t feature tagging. To get around this issue, we use different algorithms of computer vision to describe photos, before forwarding this information to our search engine.

How has the industry response been to your solution?

The response from the industry has been very positive, and we have been featured in several important industry applications such as Techcrunch, the Next Web and AdWeek. There is a real appetite for something different from run-of-the-mill stock photography, and this has been evident in Lobster’s media coverage.

What excites you most about the future of deep learning? Where do you hope the technology is going / where do you see the greatest opportunities?

The nature of our project means we are interested in deep models of computer vision. There are interesting approaches to describing the content of images, detecting objects and making classification. We want to create a system for selecting images using AI without the need for human input.

How do you keep up with new papers and publications?

We learn new industry trends through scientific papers, subscribing to profile mailings and researching the web for information.

What do you wish you had known before getting into deep learning?

We have encountered a great deal of difficulty with marked image datasets for training our own detectors and classifiers.