As 2022 comes to a close, and we look forward to 2023, we wanted to take a few minutes to reflect on an eventful and memorable year while celebrating some highlights! Thanks to the ongoing and unwavering support of our customers, investors, partners, and team members, we hit a few significant milestones - chiefly securing Series B funding and growing to over 100 employees across the globe.
Throughout this year, we also released some powerful new features, including new annotation types, expanded auto-labeling support, a dedicated manual QA process with reviewer roles, and much more. Even more exciting is that we have much bigger releases coming in the first half of 2023 (hint: more powerful and easy-to-use automation + DataOps).
All these events, and then some, have contributed to our ongoing mission of making AI and computer vision more accessible for companies of all shapes and sizes. We’re incredibly proud of everything our team has accomplished this year and are more inspired than ever by our customers and the innovative ways they apply CV to real-world problems.
Quick highlights from across Superb AI
Series B: We announced our $16M Series B funding! This was led by Korea Development Bank (KDB) and Series A investor Premier Partner. Seed and Series A investors Duke University (Duke Angel Network) and KT Investment also provided additional funding.
Team Growth: From a headcount perspective, we grew many times over to 100 employees! We even opened our third global office, this time in Japan.
Data and Product Growth: Over 130M labels, consisting of 450M annotations across 10K+ projects, have now been created using the Superb AI Suite. As our team has become increasingly multinational, we also wanted our platform to be the same. We fully localized our platform and supporting documentation in Japanese to facilitate this. Korean localization is on our plate for 2023!
New Newsletters: We published three newsletters this last year, two for the US market and one for the Korean market, resulting in 39 total editions and over 16K unique views. This includes MLOps Insight, which collects news and insights related to MLOps in Korea; the Ground Truth, a community-focused newsletter for computer vision practitioners in the US and abroad; and our general company newsletter, which showcases the latest Superb AI product developments and practical how-to content**.**
Conference Participation: We attended the industry’s leading artificial intelligence conference Ai4 2022, the Computer Vision and Pattern Recognition Conference (CVPR) 2022, and others, such as the Data + AI Summit, Computer Vision Summit, and apply(conf). We had many intriguing discussions and built some great relationships!
Speaking Opportunities: I, and the wider team, participated as guest speakers at more than 13 events and engagements. Among these were the AWS Seoul Summit 2022, AWS Community Day 2022, Computer Vision Summit, AIAI CV Summit, MLOps World Bay Area Summit, and more**.**
Media Coverage: We were selected as a “Korea AI Startup 100” for the second year in a row and were awarded the Presidential Citation in the 'Information and Communication Development' category**.**
Partnerships: We’ve been involved with the AI Infrastructure Alliance for a while now, contributed to the AI Infrastructure Ecosystem Report, and spoke at the Data-Centric AI Summit.
A visual look at 2022
Creating a place for more accessible AI
This year, we focused heavily on releasing features to help data labeling and engineering teams more effectively manage the data preparation process using AI and build scalable data pipelines for computer vision applications of all stripes. If you didn’t have the time to read all our blog posts about what we’ve shipped this year, now’s your chance to play a little catch-up.
DataOps Beta: Curating, labeling, and consuming the right data is often more critical than getting more data. That’s why we’ve been hard at work building DataOps, which is set for release this coming year. Here’s a quick recap of some (but certainly not all!) of its major features. First is mislabel detection, which uses novel AI to uncover labeled data that is likely misclassified. Next is semantic search, which provides a flexible and intuitive way to explore and search your dataset using natural language or image queries. And lastly is automated test and training set curation, which allows you to create ideal and realistically distributed datasets effortlessly by eliminating ad-hoc data selection practices.
Auto-Labeling Improvements: We’ve been hard at work improving the effectiveness and accuracy of our auto-labeling models. So far, we’ve improved accuracy by 2.6x and labeling speed/efficiency by 2x. More on these updates in the new year!
Near Real-Time Analytics: We replaced our batch processing pipeline with a near real-time event-driven pipeline, so your analytics dashboards update in seconds, and reports are generated in less than 10 seconds (scaling to less than 1 minute for 500,000 labels).
Dedicated Reviewer Role: Establishing an effective and structured quality assurance workflow is critical when it comes to both data quality and labeling scalability. That’s why we released a dedicated reviewer role, separate from managers, to focus solely on critical audit-related tasks. We’ve also updated our analytics dashboard, so you can drill down into and examine the effectiveness of both your labeling and QA workflows and teams**.**
New Annotation Types: We wanted to provide a better way for computer vision models to interpret objects captured with a 2d camera lens, so we added support for annotating 3d objects in 2d data using 2d cuboid annotation.
New Cloud Storage Integrations: Lastly, we added support for Microsoft’s object storage service, Azure Blob Storage, which functions similarly to our other cloud storage integrations, including optional read-only access for added security**.**
Focusing on product excellence
Of course, as much as we love building new features, big and small, 2022 was also the year we learned to slow down to focus on making our product even easier to look at and use. Here are a few highlights:
New onboarding tutorials
Improved page and app usability
New sync button for analytics dashboard
Native support for Japanese, with Korean in-development
Looking forward to 2023: DataOps
Too often, as practitioners, we default to following our intuition when it comes to data distribution, and we spend more time on improving our code or finding a better algorithm than we do on improving the data we already either have or need to collect. However, the bad data remains uncorrected when we do this, leading to more data redundancy or bias. In addition, as the size and variety of your training data increases, the probability of error and imbalance increases exponentially, making it even more difficult to improve model performance.
That’s why we’ve been working on DataOps - to give your machine learning team consistent access to high-quality data. Stay tuned for more updates early in 2023!
Helping labeling and ML teams build better workflows
We consider ourselves an extension of our customers, so we’ve focused our efforts on producing practical resources to support labeling and ML teams in improving efficiencies and reducing time-to-label, including best practice guides and how-to’s we’ve learned from working with thousands of teams like yours.
This whitepaper, written by the Superb AI team, provides a step-by-step understanding of how to effectively curate your data and automate the process using the platform.
The team worked on a new series of blogs that enable labeling, data, and machine learning teams to create ideal workflows by teaching them how to get the most out of the Superb AI platform and tech. Part 1 focuses on project setup, Part 2 on labeling your dataset, and the recently released Part 3 on building awesome ground truth datasets and establishing a more effective QA process. Be on the lookout for more to come!
We’ve identified some strategies and best practices you can employ to simplify your image annotation efforts. Also, check out similar articles we’ve written for video annotation and building ideal review and QA workflows.
Finally, we understand that not everyone already has a team for labeling datasets. So we put together some articles on effectively finding and engaging annotation services that can do the hard work for you.
See you in 2023
Can you believe it’s already 2023? This past year has flown by! We’re constantly amazed at how quickly the Superb AI product and community have grown, and we’re continually grateful for everyone that has supported us up to this point.
This year, we’ll invest heavily in bringing major new products and features to market and building the foundation for the AI development ecosystem, with the aim to introduce AI to all companies someday. This includes novel applications of AI to help you focus on and improve data curation and data quality, new methods of annotation automation, and much more. And, of course, we’ll continue to make enhancements to our labeling platform. We can’t wait to show you everything, so stay tuned!
To help us on our mission to automate data preparation, freeing teams to focus on the work that matters most and thereby make computer vision more accessible for all, check out our open positions on our careers page. Or join the conversation with other community members on LinkedIn and Twitter.