This Week in 3D Imaging



3D Imaging Research

Stanford University researchers tackled the issue of 3D reconstruction from a single image, producing a simple form of output in the form of point cloud coordinates. They developed an innovative and efficient architecture, loss function, and learning paradigm in response to this unconventional output form and the inherent uncertainty in the groundtruth. The team's final design is a conditional shape sampler that can anticipate various realistic 3D point clouds from an input image. In tests, their system not only outperforms cutting-edge techniques on benchmarks for single image-based 3D reconstruction, but it also exhibits strong performance for 3D shape completeness and shows promise in terms of its capacity to make several credible predictions.

MIT 3D Imaging

Three different contributions were made by MIT researchers that studied 3D shape modeling from a single photograph. They started by introducing Pix3D, a big benchmark of various image-shape combinations with pixel-level 2D-3D alignment. Pix3D is widely used for tasks involving shapes, such as reconstruction, retrieval, viewpoint estimation, etc. Secondly, they calibrated the assessment criteria for 3D shape reconstruction through behavioral experiments in order to objectively and methodically compare cutting-edge reconstruction algorithms on Pix3D. Third, they created a novel model that does both position estimation and 3D reconstruction at the same time, and their multi-task learning strategy achieves cutting-edge performance on both tasks.

Open Source:

Open Source AI News

The open-source artificial intelligence model StarCoder, which can produce code in a variety of programming languages, was introduced by ServiceNow Inc. and Hugging Face Inc. In the open-source community, StarCoder, according to the companies, is the most sophisticated model of its kind. It was created as a result of research that ServiceNow and Hugging Face began last year. The developers from the two firms as well as hundreds of other AI specialists all contributed to the initiative, which is termed BigCode.


A growing open source movement is being backed by numerous businesses, including hardware manufacturers, to guarantee that AI technologies are affordable and available. By restricting access to the tools that are defining the AI environment, tech giants like Google, Microsoft, and Facebook are now attempting to cash in on the AI gold rush. Access to technologies that can produce graphics, respond to queries, and comprehend spoken languages is being restricted. But for some, particularly hardware manufacturers, ceding control of AI to a small group of wealthy tech firms is bad for business. The emergence of open source is being supported by some businesses, making AI technologies affordable and available.

Register for a Free Trial of Checkpoint Today!