Samsung Invests in Video AI Startup - Analyzing Footage at Scale

The Challenge of Long-Form Video Analysis with AI

Presently, numerous artificial intelligence systems possess the capability to analyze video content and generate summaries of the events depicted. However, a significant hurdle arises when these models are tasked with processing inquiries concerning multiple videos, or footage extending over considerable durations.

This limitation poses a substantial problem for security firms seeking to leverage AI for the efficient review of extensive footage from diverse camera sources. Similarly, marketing organizations aiming to evaluate various video campaigns and product demonstrations encounter difficulties.

Memories.ai: A Platform for Extensive Video Processing

Memories.ai is developing an AI platform designed to address this challenge, with the capacity to process up to 10 million hours of video data. The startup intends to deliver a contextual framework for organizations managing large video archives, incorporating features like searchable indexing, tagging, segmentation, and data aggregation.

The company’s co-founder, Dr. Shawn Shen, previously served as a research scientist at Meta’s Reality Labs during his doctoral studies. His colleague, Enmin (Ben) Zhou, held a position as a machine learning engineer at Meta.

The Focus on Contextual Understanding

“Leading AI companies, including Google, OpenAI, and Meta, are primarily focused on developing end-to-end models. While these models offer valuable capabilities, they often exhibit limitations in comprehending video context beyond a few hours,” explained Shen to TechCrunch.

“However, human visual memory relies on processing a broad context of information. We drew inspiration from this and aimed to create a solution that enhances the understanding of video content across extended timeframes,” he added.

Seed Funding and Investor Perspectives

To further this development, the company has secured $8 million in seed funding, led by Susa Ventures, with participation from Samsung Next, Fusion Fund, Crane Ventures, Seedcamp, and Creator Ventures. Initially targeting a $4 million raise, Memories.ai received an oversubscribed investment due to strong investor interest.

Misha Gordon-Rowe, a partner at Susa Ventures, commented, “Dr. Shen is a highly skilled technical founder, deeply committed to advancing the boundaries of video understanding and intelligence. Memories.ai has the potential to unlock significant first-party visual intelligence data with its solution. We identified a market gap in long-context visual intelligence, which motivated our investment.”

Samsung Next’s investment thesis centered on the potential applicability of Memories.ai’s solution to consumer applications.

“A key aspect of Memories.ai that appealed to us is its ability to perform substantial on-device computing. This reduces the necessity of storing video data in the cloud, potentially enhancing security for individuals concerned about privacy when deploying security cameras,” stated Sam Campbell, a partner at Samsung Next.

Technical Approach and Core Features

Memories.ai utilizes a proprietary technology stack and models for its analyses. The process begins with noise reduction in videos, followed by compression to retain only essential data. An indexing layer then enables searchable access to the video data using natural language queries, alongside segmentation and tagging features.

Finally, an aggregation layer summarizes data from the index, facilitating the creation of comprehensive reports.

Target Markets: Marketing and Security

Currently, the startup serves two primary sectors: marketing and security. Marketing firms can leverage the platform to identify trends related to their brands on social media and determine optimal video content strategies. Memories.ai also provides tools to assist marketers in video creation.

The company is also collaborating with security organizations to analyze security footage, identifying potentially hazardous actions through pattern recognition.

Future Development and Vision

At present, organizations utilizing Memories.ai must upload their video libraries to the platform for analysis. However, Shen indicated plans to enable seamless content synchronization through shared drives in the future.

The goal is to allow users to pose questions such as: “Provide a summary of all interviews conducted in the past week.”

Shen envisions an AI assistant capable of contextualizing a user’s experiences through photos or activation of smart glasses. He also foresees the technology contributing to the training of humanoid robots for complex tasks and enhancing the memory capabilities of self-driving vehicles.

Team Expansion and Competitive Landscape

The company currently has a team of 15 employees and intends to utilize the funding to expand its workforce and refine its search capabilities.

Memories.ai faces competition from startups like mem0 and Letta, which are focused on creating memory layers for AI models, although their current video support is limited. It also competes with established players such as TwelveLabs and Google, who are actively developing AI solutions for video understanding.

However, Shen believes his company’s solution is more versatile, allowing it to integrate with various video models.