首页|New Findings from Indian Institute for Technology in the Area of Robotics Report ed (Physically Plausible 3d Human-scene Reconstruction From Monocular Rgb Image Using an Adversarial Learning Approach)

New Findings from Indian Institute for Technology in the Area of Robotics Report ed (Physically Plausible 3d Human-scene Reconstruction From Monocular Rgb Image Using an Adversarial Learning Approach)

扫码查看
By a News Reporter-Staff News Editor at Robotics & Machine Learning Daily News Daily News-New research on Robotics is the subjec t of a report. According to news reporting from Maharashtra, India, by NewsRx jo urnalists, research stated, "Holistic 3D human-scene reconstruction is a crucial and emerging research area in robot perception. A key challenge in holistic 3D human-scene reconstruction is to generate a physically plausible 3D scene from a single monocular RGB image." Financial supporters for this research include Australian Research Council, US A ir Force Research Laboratory, Defense Advanced Research Projects Agency (DARPA). The news correspondents obtained a quote from the research from Indian Institute for Technology, "The existing research mainly proposes optimization-based appro aches for reconstructing the scene from a sequence of RGB frames with explicitly defined physical laws and constraints between different scene elements (humans and objects). However, it is hard to explicitly define and model every physical law in every scenario. This letter proposes using an implicit feature representa tion of the scene elements to distinguish a physically plausible alignment of hu mans and objects from an implausible one. We propose using a graph-based holisti c representation with an encoded physical representation of the scene to analyze the human-object and object-object interactions within the scene. Using this gr aphical representation, we adversarially train our model to learn the feasible a lignments of the scene elements from the training data itself without explicitly defining the laws and constraints between them. Unlike the existing inferencet ime optimization-based approaches, we use this adversarially trained model to pr oduce a per-frame 3D reconstruction of the scene that abides by the physical law s and constraints. Our learning-based method achieves comparable 3D reconstructi on quality to existing optimization-based holistic humanscene reconstruction me thods and does not need inference time optimization."

MaharashtraIndiaAsiaEmerging Techn ologiesMachine LearningRobotRoboticsIndian Institute for Technology

2024

Robotics & Machine Learning Daily News

Robotics & Machine Learning Daily News

ISSN:
年,卷(期):2024.(Mar.11)