This Week in 3d Imaging

Research Highlights:

A truly differentiable rendering framework was proposed by researchers from the University of Southern California. It is capable of rendering colorized meshes directly using differentiable functions and back-propogating effective supervision signals to mesh vertices and their attributes from a variety of image representations, including silhouette, shading, and color images. A innovative formulation that sees rendering as an aggregation function that merges the probabilistic contributions of all mesh triangles with respect to the displayed pixels is the foundation of their system.

From a single monocular photograph, University of Pennsylvania researchers built a 3D model of the human torso, hands, and facial expression. To do this, they train SMPL-X, an extension of SMPL that adds fully articulated hands and an expressive face, using tens of thousands of 3D scans.

Open Source Highlights:

Password management company?Dashlane?has made its mobile app code available on GitHub for public perusal, a first step it says in a broader push to make its platform more transparent. The Dashlane Android app code is available now?alongside?the iOS incarnation, though it also appears to include the codebase for its Apple Watch and Mac apps even though Dashlane hasn’t specifically announced that. The company said that it eventually plans to make the code for its web extension available on GitHub as well.

The Indian startup Toolset announced today that it has raised $4.6 million in a pre-Series A round of funding to capitalize on the growing popularity of an open source project it launched 20 months ago. ToolJet, essentially, allows companies to create custom internal business applications, such as admin panels or order tracking systems, by connecting to data sources such as Postgres, MySQL or Airtable — minimal coding required, with a visual-based front-end builder in tow.