Researchers from Nanjing University of Science and Technology presented a photon counting 3D imaging technique that overcomes the challenge of scaling to high spatial resolution. The technique combines short-pulsed structured illumination and a single-pixel photon counting detector. It utilizes multi-resolution photon counting and Hadamard multiplexing along wavelet trees to acquire a high-resolution 3D image from a coarse image and edges sampled at finer resolutions. The use of Hadamard multiplexing significantly increases the detected power. By performing wavelet-tree-based edge prediction and Hadamard demultiplexing, both the required measurements and reconstruction time are greatly reduced, making the technique suitable for high-resolution scenes. Experimental results show that a 3D image with a resolution of up to 512 × 512 pixels can be acquired and retrieved in as little as 17 seconds.
A simple computational camera for single-shot 3D imaging was demonstrated by researchers from the University of California. Their lensless setup was comprised of a diffuser in front of an ordinary image sensor. They efficiently address the large-scale inverse problem by means of a physical approximation and a straightforward calibration scheme. 100 million voxels are rebuilt from a single 1.3 megapixel image using their 3D voxel grid, which was selected to match the experimentally measured two-point optical resolution over the field-of-view. Nevertheless, the effective resolution varies greatly depending on the scene material. They offer novel theories for analyzing resolution in such systems because this effect is common to a wide variety of computational cameras.
Open Source News:
Abu Dhabi has released the source code of its large language model, Falcon 40B, inviting developers from academia and the private sector to build new AI applications. This move, which separates Abu Dhabi's Technology Innovation Institute from its competitors, is causing division in the industry as it freely gives away AI technology to be modified by skilled coders. While Google and OpenAI have chosen to keep their foundational models closed, expressing concerns about the potential misuse of large language models (LLMs), Abu Dhabi's open-source approach is seen by proponents of open source software as a crucial step towards fostering AI innovation. Making Falcon 40B open source allows researchers and entrepreneurs to explore the most innovative use cases, promoting transparency, accountability, and further advancements in the field.
Ballerine, an open-source risk decisioning platform, has secured $5 million in seed funding led by Team8. The company offers a platform that integrates global data sources and provides tools for automating and improving decision-making processes in areas such as onboarding, underwriting, and transaction monitoring. Ballerine serves businesses subject to regulations like KYB, KYC, and anti-money laundering, including financial institutions, fintech companies, e-commerce platforms, and more. It has gained significant traction, with over 1500 developers worldwide using its code and several fintech companies implementing it in production. Ballerine's platform simplifies the complex tasks of verifying, underwriting, and monitoring businesses. It allows developers to customize their risk assessment workflows through a marketplace of point solutions, accessing data sources through a single agreement with Ballerine or directly with data vendors. By leveraging dispersed data, streamlining privacy management, and offering control and flexibility, Ballerine enables companies to enhance their risk decisioning processes while maintaining control. The open-source core of Ballerine's platform gives developers and risk managers more control over infrastructure and allows customization to specific needs, catering to global financial services and large institutions with various use cases.