Computer Vision News 2 NeurIPS 2023 Accepted Paper Ayca Takmaz (left) is a second-year PhD student at ETH Zurich, under the supervision of Prof. Bob Sumner and Prof. Siyu Tang. Ayca also closely collaborates with Dr. Francis Engelmann. In her research, she focuses on 3D scene understanding, particularly for augmented reality applications. Elisabetta Fedele (right) is a Master’s student at ETH Zurich Computer Science department, specializing in theoretical computer science and machine learning, with a focus on computer vision. by Ayca Takmaz and Elisabetta Fedele 3D instance segmentation, which is the task of predicting 3D object instance masks along with their object categories, has many crucial applications in fields such as robotics and augmented reality. In recent years, 3D instance segmentation approaches have achieved significant success. However, current methods operate under a closed-set paradigm - they can typically only recognize object categories from a pre-defined closed set of classes that are annotated in the training datasets. As a result, these methods often have a limited ability to understand a scene beyond the categories seen during training, and cannot directly answer more natural, free-form questions such as “Where can I sit?” “Where is the side table with flowers on it?”, “Do you know where my leather bag is?”, “Where are my keys, the ones with the blue keychain?" or identify less common objects such as “a toy penguin”, “Cetaphil soap”, and “an abstract-style wall-art”. OpenMask3D: Open-Vocabulary 3D Instance Segmentation
RkJQdWJsaXNoZXIy NTc3NzU=