Memo - Roboflow

Published on Feb 1, 2021

Company Overview

Roboflow is a feature-rich platform for working with computer vision ML models. Roboflow makes it possible for non computer-vision specialists to easily host datasets, train new ML models on common architectures, and deploy them in a variety of formats.

With the proliferation of ever more powerful cameras, new machine learning techniques, and the ability to train and deploy computer-vision based ML models in both the cloud and on the edge; computer vision is a field that is only going to continue to grow in its importance across a variety of sectors. Some of these include:

  • Agriculture

  • Security

  • HealthCare and Medicine

  • Games

  • Mobile Apps

  • Robotics

  • Augmented Reality

Computer Vision and Machine Learning are complex technical areas, with lots of tooling that takes a long time to get properly configured. Developers often have to write lots of boiler-plate code for working with image datasets, as well as spending lots of time getting their datasets ready for training through augmentations and labelling. Roboflow helps with these processes - by making computer-vision based machine learning workflows easier, it lets developers focus more time on writing their applications, rather than going through the tedious processes of setting up their own Computer Vision processing pipeline.

Roboflow’s expertise in computer-vision and machine learning, combined with their rapid execution ability, make it well positioned to take advantage of the computer vision revolution.

Product

Developers can easily create new ML models based off of common architectures, such as YOLO5, MobileNet, and ResNet. These models can then be exported in a variety of formats that are common to standard machine learning workflows.

Roboflow also includes tools for pre-processing datasets, eliminating the need for writing common boilerplate computer vision code. It also includes tools for determining the health of a dataset, as the ability to host. A relatively new feature, Roboflow Annotate, allows for annotating datasets in Roboflow, but Roboflow also provides integrations for other common labelling services and tools. Their latest feature is an API that allows developers to directly upload new images for training directly from an application.

Roboflow enables developers without deep machine learning expertise to create vision models, similar to how software engineers do not need to know cryptography to accept payments.

Pricing

Roboflow has three tiers of pricing:

  • Starter - For individuals and small companies experimenting with new computer vision applications

  • Professional - Enhanced features, including more more dataset analysis and augmentation features; and collaboration for teams

  • Enterprise - Higher-grade security, dedicated support, unlimited teams and sharing

Growth / Adoption

Roboflow has had over 10,000 users since their January release ranging from individual developers to startups and Fortune 50 companies.

Roboflow has already been used in supply chain optimization, gaming, traffic optimization, environmental improvement, cancer research, accessibility, and much more.

Founders and Team

Roboflow was founded by Joseph Nelson and Brad Dwyer in January 2020. They first met each other in 2000. Roboflow is the second startup that Joseph has founded; his previous one was acquired in 2018. It is also Brad’s second startup - his last company, in the social games space, reached 10 million players and is an active ten-person company that Brad left to start Roboflow.

There are three full-time team members, in addition to the founders; and they are looking to bring on another full-stack engineer very soon. Additionally, they have on intern working with them.

The team is fully remote, with Joseph based in Washington D.C., and Brad based in Des Moines, Iowa. The whole team did spend the summer of 2020 working together in a hacker house in Des Moines while they were going through YC. All of the current team members have either worked together before, or have known each other prior to working at Roboflow.

I’ve used Roboflow on some experiments I was working on earlier this year, and found the team to be extremely responsive to bugs and feature requests. They ship new features at an impressive clip, too.

My impression of the founders are that they are exceptionally hard working and earnest, and want to build a company that pushes the envelope in terms of working with computer vision; while also being one that will attract exceptional talent.

Investment Details

Roboflow participated in the YC summer 2020 session. They recently announced a $2.1M Seed Round, led by Lachy Groom and Craft Ventures, and are currently hiring for several roles.