Samsung supports an AI video startup which can analyze thousands of hours of sequences

Many AI tools can consult a video today and summarize what is going on, but things become a little delicate when asking questions about the models on several videos and images covering several hours.

This is a great limitation for security companies who wish to use AI to rub thousands of hours of sequences of different cameras, as well as marketing companies that wish to study different video campaigns and products of products.

Souvenirs.ai wishes to solve this problem with its AI platform which can deal with up to 10 million hours of video. For companies with many videos to analyze, the startup wishes to provide a contextual layer, with indexing, marking, segments and aggregation available.

His co-founder, Dr. Shawn Shen, was a researcher with Meta’s reality laboratories when he was continuing his doctorate, and his counterpart in min (Ben) Zhou worked at Meta as an automatic learning engineer.

“All the best companies in AI, such as Google, Openai and Meta, are focused on the production of end-to-end models. These capacities are good, but these models often have limits concerning the understanding of the video context beyond one or two hours,” Shen told Techcrunch.

“But when humans use visual memory, we are scrutinizing a great context of data. We were inspired by this and we wanted to build a solution to understand the video through several hours,” he said.

Towards this objective, the company has now raised $ 8 million in a seed financing cycle led by Susa Ventures, and with the participation of Samsung Next, Fusion Fund, Crane Ventures, Seedcamp and Creator Ventures. Shen said that the company was initially aimed at raising $ 4 million, but ended up with a jerky round due to the interest of investors.

“Shen is a highly technical founder, and he is obsessed with the push of the limits of understanding and video intelligence,” said Misha Gordon-Rowe, partner of Susa Ventures. “Memories.Ai can unlock a lot of first visual intelligence data with its solution. We estimated that there was a gap on the market for a long context of visual intelligence, which attracted us to invest in the company,” he added.

Samsung then had a slightly different thesis: Samsung’s investment branch sees the memories solution .i is useful for consumers.

“One thing we love in memories is that it could do a lot of information on the devices. This means that you don’t necessarily need to store video data in the cloud.

Memories.ai says he uses his own technological battery and models to carry out analyzes. First of all, it deletes the noise from the videos and passes the output via a compression layer to store only important data. Then there is an indexing layer, which makes the video data available (using natural language queries) with segmentation and beacons. There is also a layer of aggregation that summarizes index data, helping to create reports.

Currently, the startup is aimed at two types of businesses: marketing and security. Marketing companies can use startup tools to search for trends related to their brands on social networks and identify the type of video they want to make. Memories .i also provides tools to marketing specialists to create these videos.

The company also works with security companies to help them analyze security images to determine the potentially dangerous actions of people in videos by reasoning through models.

Currently, companies working with Memories.ai must download their video library on the platform to analyze clips. But Shen said that in the future, its customers will be able to more easily create a reader and shared synchronization reader. The plan is to allow customers to ask questions such as: “Tell me about the people I interviewed last week.”

Shen is considering an AI assistant who can gain context on the life of a user via his photos or when he activated smart glasses. He also sees technology playing a role in the training of humanoid robots to perform complex tasks or helping autonomous cars remember the different ways.

The company currently employs 15 people and plans to use the fund to increase its team and improve its research.

Memories.a is against similar startups, such as MeM0 And ReadWho work to provide a layer of memory for AI models, although they currently offer limited video support. He also has to deal with companies like Twelvelabs and Google, who have worked to help AI models to understand the videos.

Shen, however, believes that the solution of his business is more horizontal, which would also allow it to work with different video models.