Adapt @ CVPR

Adapt’s Generative AI recruitment specialists, Kieran Toner (Director) and Maisie Broym (Principal Consultant) attended this years CVPR event in Seattle, so we caught up with them to hear their take on the current market and insights from other professionals at the event.

Where were the LiDAR’s this year?

We saw some options for sensor based cameras and some cool VR options but compared to the previous years, a complete lack for ViDAR.

Hello, Annotation Data

Increase in presence of Data Annotation companies, this year which demonstrates an integration with the wider ecosystem, as previously CVPR was a research only event.

Focus on Large Vision Models

This year there was a big focus on large vision models which we think has gained more momentum as a result of the recent release of GPT -4o and GPT -4. However, one attendee said they think, “the computer vision generative AI hasn’t had it’s ‘LLM moment’ yet, likely because of tokenization & lack of fine control.” They also noted, that “The industry exhibition hall at the event was overwhelmed by annotation, data discovery and curation tools.”

Closer tech partnership for Seoul & the US?

“CVPR 2024 had the feel of a global conference. Specifically, I was impressed with the research papers and start-ups from South Korea… I foresee a closer partnership between South Korea and the US in the coming years.” Said Kieran.

What does the increase in AV players mean for Recruitment?

“There was a large presence of AV players, for example Latitude AI and Gatik were exhibiting the newest gen vehicles which indicates a play in to LLM’s and Generative AI development for their newest vehicles. This is already being showcased by companies in the space developing software such as Wayve, Helm and Waabi. From a recruitment perspective this changes the playing field. Whilst previously hiring from competing companies was the target, it is now essential for these companies to hire from outside of the space to ensure they can innovate ahead of other competitors.” Said Maisie

New players in Facial Recognition

Maisie also said, “A lot of interesting players in facial recognition and mapping/masking technology for multiple application areas – L’Oreal was present and showcasing their facial recognition tech.”

Manipulating Embedding Spaces

A tech professional in attendance made the observation that “this years main theme was embedding spaces. The event showcased clever methods for controlling the output of foundation models in Generative AI. Most research groups are limited to a few GPUs, therefore, optimization and efficiency are obvious research areas.”

Formula for Synthetic Data Generation

One developer we spoke to, told us there’s a rule of thumb when using synthetic data: “Never go over 3 parts Synthetic and 1 part Real” They said, “it’s unclear why today’s best synthetic data can not achieve the same performance as real data, nevertheless its something to be aware of.”

Funding for Autonomous Driving

One candidate observed that “there was a lot of interest and funding for autonomous driving this year, and although autonomous driving has been at previous CVPR events, it felt much more serious and realistic than a few years ago.”

The with key players in the industry showcasing their latest technologies and highly skilled professionals speaking and in attendance, the CVPR event is always a great indication of how the greater tech ecosystem will develop in the upcoming year and the growth trajectory of specific areas like computer vision, embedding spaces, facial recognition and less specifically Generative AI. It seems there are exciting developments happening, not only in the US but also across the globe, in countries like South Korea.