When Julien Trombini and Guillaume Cazenave started online video-analytics startup Two-i four several years in the past, they had an bold purpose: increasing the high quality of city lifetime by one particular working day staying capable to monitor a city’s streets, garbage collection and other community services.
Together the way, the pair discovered a wholly distinct area of interest. Right now, the company’s technology — which combines computer system eyesight, knowledge science and deep learning — is helping to reduce lethal accidents in the oil and gas business, a single of the world’s most perilous sectors.
Initially, Trombini and Cazenave envisioned a program that would empower civic leaders to see what improvements were essential across a municipality.
“It would be like possessing a weather conditions map of the metropolis, only just one that actions effectiveness,” mentioned Trombini, who serves as chairman of Two-i, an NVIDIA Metropolis partner dependent in Metz, a historic metropolis in northeast France.
That proved a tall order, so the two refocused on specific facilities, these as stadiums, retirement houses and transit stations, where by its tech can help with safety and incident detection. For instance, it can inform the proper individuals when a retirement house resident falls in a corridor. Or when a transit rider working with a wheelchair simply cannot get on a prepare simply because of a broken elevate.
Extra just lately, the business was approached by ExxonMobil to support with a likely fatal challenge: improving employee basic safety about open oil tanks.
With each other with the electrical power huge, Two-i has developed an AI-enabled online video analytics application to detect when people near a hazard zone and chance slipping and straight away notify others to acquire fast motion. In its initial months of procedure, the vision AI technique prevented two mishaps from developing.
Even though this use circumstance is very specific, the company’s AI architecture is designed to flexibly support lots of distinct algorithms and capabilities.
“The algorithms are precisely the exact as what we’re applying for unique consumers,” said Trombini. “It’s the same technologies, but it’s packaged in a diverse way.”
Building the Most of Eyesight AI
Two-i’s adaptability stems from its reliance on applying the NVIDIA Metropolis system for AI-enabled video analytics apps, leveraging superior instruments and adopting a total-stack strategy.
To do so, it relies on a variety of NVIDIA-Certified Methods, making use of the newest workstation and information center GPUs centered on the large-performance NVIDIA Ampere architecture, for both coaching and inference. To shorten teaching instances additional, Two-i is searching to take a look at its huge image dataset on the potent NVIDIA A100 GPU.
The firm seems to commonly up grade its GPUs to guarantee it’s presenting buyers the speediest probable option, no make a difference how quite a few cameras are feeding facts into its technique.
“The time we can help save there is crucial, and the greater the components, the additional exact the results and more rapidly we get to industry,” claimed Trombini.
Two-i faucets the CUDA 11.1 toolkit and cuDNN eight.1 library to assist its deep studying approach, and NVIDIA TensorRT to speed up inference throughput.
Trombini claims one particular of the most compelling pieces of NVIDIA tech is the NVIDIA TAO Toolkit, which will help the business retain prices down as it tinkers with its algorithms.
“The heavier the algorithm, the a lot more high priced,” he stated. “We use the TAO toolkit to prune algorithms and make them a lot more customized to the task.”
For case in point, education that in the beginning took up to two months has been slashed to 3 days using the NVIDIA TAO Toolkit, a CLI- and Jupyter Notebook-primarily based version of the NVDIA practice, adapt and improve framework.
Two-i has also commenced benchmarking NVIDIA’s pretrained models from its algorithms and started employing the NVIDIA DeepStream SDK to greatly enhance its movie analytics pipeline.
Making on Good results
Two-i sees its means to clear up complex difficulties in a wide variety of options, these as for ExxonMobil, as a springboard to swinging back again about to its initial wise city aspirations.
By now, it’s monitoring all roadways in 8 European metropolitan areas, examining website traffic flows and knowledge in which cars and trucks are coming from and heading to.
Trombini recognizes that Two-i has to continue to keep its concentration on delivering a person profit right after a different to obtain the company’s extended-expression targets.
“It’s coming slowly,” he claimed, “but we are beginning to implement our vision.”