Periodic Reporting for period 2 - Nyris (Transforming industrial part replacement and after sales through visual search and synthetic images)
Berichtszeitraum: 2022-09-01 bis 2023-08-31
Identifying and retrieving spare parts quickly and efficiently is a growing necessity of the manufacturing industry. Our visual search engine has proven to be the most accurate tool against several large competitors that allows any spare part to be found in manufacturers’ catalogues by simply taking or sharing an image of it. Most importantly, we can overcome the key limitation in implementing visual searches for industrial applications: the availability of images for AI network training and indexing. While many images are publicly available for consumer retail, large manufacturers often cannot undertake the time-consuming and expensive effort of producing these images for all spare parts and machines through manual photography. We are the only company in the market who can produce synthetic images from CAD data, which outperform real data in search accuracy. Moreover, we can onboard customers in a matter of minutes. At Nyris, we give search the power of sight.
In the initial year of the project, we demonstrated the remarkable efficiency and scalability of our synthesization process, surpassing our data synthesization time objectives. We proved that using synthetic data to train vision AI models results in high-performance recognition networks. These networks were incorporated into our Proof of Concept (PoC) Demo Application, developed in the first year, showcasing our capabilities to potential customers.
In the project’s second year, we achieved significant milestones. Successful pilot trials were conducted, some of which have transitioned into paid licenses, marking a pivotal achievement. Our active participation in various fairs, including invitations from customers to jointly showcase our product, amplified our market presence. Our engagement with the press and research communities was robust, securing features in prominent general and specialized publications such as Tagesspiegel, Wirtschaftswoche, WDR, and VDM magazine, enhancing our visibility and engagement with our target audience.
Technologically, we attained a state-of-the-art visual search performance, applying innovative techniques like conditional latent diffusion models for synthetic data generation. Strategic steps were taken to protect our intellectual property, including filing a patent and securing the trademark "visual synonyms."
A crowning achievement was attaining ISO 27001:2022 certification. This certification is a testament to our steadfast commitment to data privacy and security, essential requirements in the industrial sector. It not only boosts the trust of our existing customers but also enhances our appeal to potential customers, underscoring our adherence to stringent data protection standards.
In the project’s second year, we achieved significant milestones. Successful pilot trials were conducted, some of which have transitioned into paid licenses, marking a pivotal achievement. Our active participation in various fairs, including invitations from customers to jointly showcase our product, amplified our market presence. Our engagement with the press and research communities was robust, securing features in prominent general and specialized publications such as Tagesspiegel, Wirtschaftswoche, WDR, and VDM magazine, enhancing our visibility and engagement with our target audience.
Technologically, we attained a state-of-the-art visual search performance, applying innovative techniques like conditional latent diffusion models for synthetic data generation. Strategic steps were taken to protect our intellectual property, including filing a patent and securing the trademark "visual synonyms."
A crowning achievement was attaining ISO 27001:2022 certification. This certification is a testament to our steadfast commitment to data privacy and security, essential requirements in the industrial sector. It not only boosts the trust of our existing customers but also enhances our appeal to potential customers, underscoring our adherence to stringent data protection standards.
Our project has yielded transformative results that redefine the boundaries of current technological capabilities. The conventional state of the art primarily relies on real images for visual search. However, our innovative approach has demonstrated that superior results can be attained through the utilization of synthetic images for index creation and vision AI development. Our synthesization pipeline stands unparalleled, manifesting exceptional speed and quality in processing CAD files. Remarkably, our pipeline is capable of processing hundreds of CAD files per minute, showcasing a phenomenal improvement in operational efficiency.
Furthermore, our synthetic data, generated through this rapid synthesization process, is enhanced with conditional latent diffusion, contributing to achieving search accuracy that is above the current state of the art. This innovative enhancement not only amplifies the accuracy of visual searches but also underscores our commitment to pushing the boundaries of technological excellence in visual search capabilities.
Through conducted study involving 55 selected candidates, we unveiled that visual search is exponentially faster and more successful compared to conventional keyword-based searches. The study revealed that visual search is 3.3 times faster and 51% more successful, marking a significant advancement in search efficiency and accuracy for spare parts.
Furthermore, our synthetic data, generated through this rapid synthesization process, is enhanced with conditional latent diffusion, contributing to achieving search accuracy that is above the current state of the art. This innovative enhancement not only amplifies the accuracy of visual searches but also underscores our commitment to pushing the boundaries of technological excellence in visual search capabilities.
Through conducted study involving 55 selected candidates, we unveiled that visual search is exponentially faster and more successful compared to conventional keyword-based searches. The study revealed that visual search is 3.3 times faster and 51% more successful, marking a significant advancement in search efficiency and accuracy for spare parts.