Search for: All records

Light’sinteraction with objectsurfacesthrough anisotropic reflection– where reflected light varies with viewing angles–offers significant potential for enhancing visual capabilities and assisting informed decision-making. Such ubiquitous light transfer phenomenon supports directional information encoding in sensing and dynamic display applications. We present LumosX, a set of techniques for encoding and decoding information through light intensity changes using 3D-printed optical anisotropic properties. By optimizing directional reflection and brightness contrasts through off-the-shelf materials and precise control over processing parameters (e.g., extrusion volume, raster angles, layer height, nozzle positioning), we enable cost-effective fabrication of visually enhanced objects. Our method supports modular assembly for highly curved regular surfaces and direct printing on top of relatively flat curved surfaces, enabling flexible information encoding for diverse applications. We showcase LumosX’s effectiveness through various indoor and smart urban sensing scenarios, demonstrating significant improvements in both human interaction and autonomous machine perception. 

Instance detection (InsDet) is a long-lasting problem in robotics and computer vision, aiming to detect object instances (predefined by some visual examples) in a cluttered scene. Despite its practical significance, its advancement is overshadowed by Object Detection, which aims to detect objects belonging to some predefined classes. One major reason is that current InsDet datasets are too small in scale by today's standards. For example, the popular InsDet dataset GMU (published in 2016) has only 23 instances, far less than COCO (80 classes), a well-known object detection dataset published in 2014. We are motivated to introduce a new InsDet dataset and protocol. First, we define a realistic setup for InsDet: training data consists of multi-view instance captures, along with diverse scene images allowing synthesizing training images by pasting instance images on them with free box annotations. Second, we release a real-world database, which contains multi-view capture of 100 object instances, and high-resolution (6k\texttimes{} 8k) testing images. Third, we extensively study baseline methods for InsDet on our dataset, analyze their performance and suggest future work. Somewhat surprisingly, using the off-the-shelf class-agnostic segmentation model (Segment Anything Model, SAM) and the self-supervised feature representation DINOv2 performs the best, achieving >10 AP better than end-to-end trained InsDet models that repurpose object detectors (e.g., FasterRCNN and RetinaNet). 

Instance detection (InsDet) is a long-lasting problem in robotics and computer vision, aiming to detect object instances (predefined by some visual examples) in a cluttered scene. Despite its practical significance, its advancement is overshadowed by Object Detection, which aims to detect objects belonging to some predefined classes. One major reason is that current InsDet datasets are too small in scale by today’s standards. For example, the popular InsDet dataset GMU (published in 2016) has only 23 instances, far less than COCO (80 classes), a well-known object detection dataset published in 2014. We are motivated to introduce a new InsDet dataset and protocol. First, we define a realistic setup for InsDet: training data consists of multi-view instance captures, along with diverse scene images allowing synthesizing training images by pasting instance images on them with free box annotations. Second, we release a real-world database, which contains multi-view capture of 100 object instances, and high-resolution (6k×8k) testing images. Third, we extensively study baseline methods for InsDet on our dataset, analyze their performance and suggest future work. Somewhat surprisingly, using the off-the-shelf class-agnostic segmentation model (Segment Anything Model, SAM) and the self-supervised feature representation DINOv2 performs the best, achieving >10 AP better than end-to-end trained InsDet models that repurpose object detectors (e.g., FasterRCNN and RetinaNet). 

The increase of distributed embedded systems has enabled pervasive sensing, actuation, and information displays across buildings and surrounding environments, yet also entreats huge cost expenditure for energy and human labor for maintenance. Our daily interactions, from opening a window to closing a drawer to twisting a doorknob, are great potential sources of energy but are often neglected. Existing commercial devices to harvest energy from these ambient sources are unaffordable, and DIY solutions are left with inaccessibility for non-experts preventing fully imbuing daily innovations in end-users. We present E3D, an end-to-end fabrication toolkit to customize self-powered smart devices at low cost. We contribute to a taxonomy of everyday kinetic activities that are potential sources of energy, a library of parametric mechanisms to harvest energy from manual operations of kinetic objects, and a holistic design system for end-user developers to capture design requirements by demonstrations then customize augmentation devices to harvest energy that meets unique lifestyle.