|
|
0 comments
Research Highlights
- A novel differentiable search framework is proposed to support fast gradient-based search within a highly flexible network topology search space. The discretization of the searched optimal continuous model in differentiable scheme may produce a sub-optimal final discrete model (discretization gap). NVIDIA researchers suggested a topological loss as a solution to this issue. The Medical Segmentation Decathlon (MSD) challenge, which consists of ten difficult segmentation tasks, is used to test their Differentiable Network Topology Search method (DiNTS). Their technique results in cutting-edge performance and the top spot on the MSD Challenge leaderboard.
- Using a combination of a fully convolutional network (FCN) and a recurrent neural network (RNN), which are responsible for taking advantage of the intra-slice and inter-slice contexts, researchers from the University of Notre Dame suggested a novel DL framework for 3D picture segmentation.
Open Source Highlights
- According to a study that was published in Radiology, an AI model accurately executes total kidney volume measurements from MRI and decreases contouring time, potentially enhancing treatment for persons with a common kind of kidney illness.
- Numerous applications, including 3D reconstruction and photorealistic synthesis of new views, are made possible by the ability of implicit representations of 3D scenes to encode hundreds of high-resolution images in a small format. Sadly, NeRFs' various flaws prohibit implicit representations from being the default 3D data format for perceptual tasks. An implicit network's training procedure, for instance, can go very slowly. The inference procedure is also too drawn out to be included into any real-time application. Last but not least, the features that a NeRF model implicitly learns are scene-specific and cannot be applied to other scenes. The aforementioned issues have recently been addressed by NeRF versions like Plenoxels. In fact, Plenoxels facilitate quick learning while preserving a constant feature representation across scenes.
Access the open-source code repository at www.github.com/aksg87/adpkd-segmentation-pytorch.