Using a single RGB image, PlaneRCNN, a deep neural network introduced by NVIDIA researchers, can identify and rebuild piecewise planar surfaces. Using segmentation masks and plane parameters, PlaneRCNN uses a variation of Mask R-CNN to find planes. Following that, PlaneRCNN jointly improves all of the segmentation masks using a unique loss that enforces the consistency with a nearby view during training. The research also introduces a new benchmark with more precise plane segmentations in the ground truth, in which PlaneRCNN beats current state-of-the-art approaches by a sizeable margin in the metrics for plane identification, segmentation, and reconstruction. Robust plane extraction is made a significant step closer with the help of PlaneRCNN, which will immediately affect several applications such as robotics, augmented reality, and virtual reality.
A unified formulation was presented by researchers from University College London for the issue of 3D human posture estimation from a single raw RGB image. This formulation considers 2D joint estimation and 3D pose reconstruction simultaneously to enhance both tasks. They used a combined strategy that combines probabilistic understanding of the 3D human position with a multi-stage CNN architecture and uses the understanding of conceivable 3D landmark locations to focus the search for more suitable 2D places. The entire procedure is trained from beginning to end, is very effective, and produces cutting-edge results on Human3.6M, exceeding earlier methods on both 2D and 3D mistakes.
Open Source News:
At the Open Source Summit North America conference , Amazon Web Services (AWS) announced it is making Cedar, a language for defining permissions as policies that includes automated reasoning to mathematically prove an IT environment is secure, available as an open source project. Furthermore, AWS released SnapChange, an open source fuzz testing tool that security researchers can use to find vulnerabilities by replaying physical memory snapshots in a KVM virtual machine. By observing the system's behavior as it processes arbitrary data, fuzz testing can identify software security flaws.
The company's PaLM 2 large language model launch is the main artificial intelligence (AI) news at Google I/O today, but it's not the only AI news. A number of open-source machine learning (ML) technology updates and improvements are also being released by the firm for the expanding TensorFlow ecosystem. TensorFlow is a Google-led open-source technology initiative that offers ML tools to assist developers in creating and refining models.