ECCV 2022 was a truly remarkable event.
With two full days of incredible workshops and tutorials, followed by three additional days of the main conference, Computer Vision was celebrated in every corner.
During the conference, giants like Amazon, Google and Meta, giants-in-the-making like Akriadata, and a whole host of awesome companies gathered to present intriguing systems, novel developments and share information.
If you couldn’t make it, we’re happy to break it down and share our insights from a wonderful week.
The expo hall was filled with exhibitors representing a diverse set of industries, ranging from Healthcare to Defense, Automotive, Manufacturing and Robotics, all who faced similar challenges and obstacles with computer vision.
Everyone of them are valiantly trying to make improvements in their domain and help everyone using various types of sensors, software and hardware. And each is trying to answer questions and find solutions with AI, models, and other various approaches.
All with something in common: The need for quality image and video data and the tools to easily obtain it.
Each company is trying to address that problem differently – whether it’s using in-house built tools or manually filtering the data, there is no hiding behind the fact that a solution is needed.
Starting with data curation and focusing on the data is as: “Data-centric AI”.
The approach has been advocated by many in the community, and the field has grown vastly since previous conferences.
Data will be the key to reaching the next phase of advancement. Training models more efficiently and improving the model’s accuracy will rely heavily on the quality of data in the novelty training sets. The more emphasis we place on the data earlier in the model lifecycle, the greater and more rapid the path to advancement will be.
“… This is a must have tool for anyone working in the field of computer vision…” E.M., ECCV 2022
The volume of visual data is a common problem across these companies, often ranging in thousands to millions of images and the volume of data only continues to grow.
Data Explorer allows data science teams to visualize the structure of the data. This, in turn, will show you where outliers lay, the different groups within your database and how to manipulate it. At the end, you could choose a diverse subset of images, that is manageable, to be used for training and testing your models.
When new batches arrive, Data Explorer allows you to visualize and compare how well the existing and the new batches match. This allows you to identify and stop drifts in the data, and in doing so, maintain high accuracy for your model and solution.
Data Explorer lets you find those diamonds buried within the mountain. Find specific images that you must have for testing or training, but are incredibly rare. Any further work on a small set of images will be cheaper and faster.
Finally, after a model training session Data Explorer will visualize the parts that need more work, and help you improve overall accuracy.
Ready to experience the power of Akridata Data Explorer? Try it out today.