Developing Apps for Drone Data

Building A Digital Brain For Drone Imagery

Ian Smith
6 min readMar 17, 2017

Humans are excellent at recognizing visual patterns and anomalies. At one point, our survival literally depended it. Quickly spotting the predator lurking amongst the bush meant the difference between life and death. Our brains and eyes have evolved over millennia just for this purpose. But what if we had to spot hundreds, or even thousands of predators simultaneously?

Fortunately, humans are also pretty good at other things, like programming.

As a software developer, you’ve embarked on a perpetual quest for optimization. For you, solving complex data problems is even more satisfying than watching this GIF:

In today’s world of flying robots, collecting terabytes of aerial imagery is quickly becoming the norm. And when businesses collect all of this data, they need to turn it into useful information — fast. Many organizations are primed to experience some pretty massive data problems, so scaling their pattern recognition abilities is a must.

DroneDeploy App Market development is cross-platform. On desktop, iOS, and Android.

This represents a massive opportunity that software developers on DroneDeploy’s App Market are helping to address.

Meet TensorFlight, a 3-founder startup that solves business problems using computer vision algorithms within a neural network. By developing an app on the App Market, TensorFlight’s entire product has been deeply integrated directly inside of DroneDeploy—and it can do some really amazing things.

The Team

Robert Kozikowski, Zbigniew Wojna, and Jan Malaszkiewicz are the co-founders of TensorFlight. They developed their professional abilities by competing in Olympiad programming and mathematics competitions when younger. Later in life, the trio met in Warsaw. From there, they split up and each went their own way, criss-crossing through tenures at Google and Facebook. During this time, they harvested valuable experience, honing their skills in machine learning and computer vision while working on some of the most popular products on the planet. In search of other challenges, they eventually realized that to achieve true fulfillment, they needed to build something of their own.

From left to right, Robert Kozikowski, Zbigniew Wojna, and Jan Malaszkiewicz

The three sought to create a company and ended up pursuing something quite exciting by building their own neural network for drone data.

The Company

After learning of the FAA’s Part 107 rule (which finally allowed businesses in the U.S. to operate drones commercially) the crew identified the enormous potential of growth for the drone market. But it wasn’t until speaking with DroneDeploy and learning about the upcoming launch of their App Market that their vision truly began to take shape. In late 2016 the team chose Mountain View, California as the headquarters of TensorFlight, Silicon Valley’s newest tech company. They immediately began building their app, which would be available exclusively on the DroneDeploy App Market.

The App

For orchards and forestry companies, knowing how many trees they’re working with is vital. Some operations manage tens of thousands of trees, spread across as many acres of land. Properly forecasting yields and documenting inventory are two of the main use cases for these organizations, so workers in the field will drive around and individually count each tree, an arduous process.

Once drones came into the picture, it suddenly became a bit easier. Instead of driving and counting on the ground, companies could quickly generate aerial imagery and then have someone count the trees on a computer screen. A quicker method, sure, but when extrapolated across long periods of time and hundreds of thousands—or even millions—of trees, still is not ideal.

A sample TensorFlight report

TensorFlight’s app solves this problem by connecting DroneDeploy users directly to their neural network which uses artificial intelligence to automatically counts the trees found in drone imagery, no human required.

Boxes signify objects that were identified

The Process

Once a customer adds the TensorFlight app to their DroneDeploy account, a new menu is presented to them when viewing their maps. This menu is essentially an iframe which developers on DroneDeploy can build within. TensorFlight’s allows the user to seamlessly send their map to their neural network by clicking a single button. This is when the magic happens.

TensorFlight’s app in the DroneDeploy UI

Behind-the-scenes, TensorFlight’s algorithms get to work. Their servers break down the map, tile by tile, and analyze the imagery using computer vision. This process is accomplished by calling DroneDeploy’s Tiles API. Then, TensorFlight’s app reports back an automated count of trees. And not only that, their app can even tell you if the tree is alive or dead.

TensorFlight’s algorithms will soon identify cars, livestock, and more

“…when comparing manual scouting to TensorFlight’s automatic tree counting, we found the automatic methods to be within 98% of the manual count. It also detected dead trees.” — David Behr, AgroDrone

TensorFlight launched their app on the DroneDeploy App Market on November 4th, 2016 and have seen tremendous growth since.

In the 4 months since launching their app on DroneDeploy, TensorFlight has been used over 800 times by more than 400 different users. Their neural network has counted 1.3 million living trees and identified 100 thousand dead trees — an average count of 12,000 trees every single day.

DroneDeploy’s ecosystem has enabled TensorFlight to connect with users across 6 continents in 4 months

Moving forward, the team is adding in even more objects to be identified, like cars and livestock.

The DroneDeploy App Market

TensorFlight’s app is available exclusively on DroneDeploy, via the App Market.

These are the three big reasons why the company chose to build their app on DroneDeploy:

  1. Distribution. More than 10 million acres of aerial data have been processed through DroneDeploy and tens of thousands of active users populate the platform. They can be found in over 150 countries and come from industries like construction, agriculture, surveying, and mining. These ingredients create the largest, most diverse set of drone data in the world.
  2. Flexibility. Code templates, design assets, and getting started videos are readily available in the developer API docs. The ability to build their app using DroneDeploy’s powerful APIs allowed them to deploy a market-ready solution in record time, launching directly into the App Market’s robust ecosystem.
  3. Monetization. Developers have full control over the price of their apps. Whether they’re free and unlimited or have an associated cost, a simple revenue share system enables developers to build a thriving business on DroneDeploy.

The development process is easy — we have our iframe within the DroneDeploy interface. Their set of easy-to-use API calls lets us access and augment user’s data within DroneDeploy. The App Market brought us more users than we expected. — Robert Kozikowski, TensorFlight CEO

The Opportunity

TensorFlight isn’t the only one. Join dozens of others on the App Market, like AirMap, Autodesk, Verifly, WhiteClouds, Skyward, John Deere, EZRoof, and many more.

Thirty percent of DroneDeploy’s thousands of monthly active mappers have added 2 or more apps to their accounts.

Using our quick templates and a bit of HTML and JavaScript, you can start creating valuable apps for the global businesses who use DroneDeploy to power their drone operations right now.

Want to build the next great app on DroneDeploy? Would your customers benefit from an integration with the world’s largest set of drone data? The next great app could be yours.

Visit to get started now.

To learn more about TensorFlight’s software, watch this overview video and read the tutorial.



Ian Smith

CEO of @WareRobotics. Ware deploys fleets of self-flying drones and machine learning inside warehouses to automate their inventory tracking.