Abstract: A physics-based network model is trained to learn weights such as trapping, detrapping, and/or transport of holes and/or electrons, as well as voltage distribution on a voxel-by-voxel basis throughout a solid-state detector model. The physics-based network may be used to estimate material property variation throughout the voxels. To reduce the number of experimental setups and information needed to train the models, the models may be trained using more easily acquired ground truth. Just the electrode signals or just the free charge data is used to train the model to characterize the solid-state detector. With this reduced data, the detector may be characterized using equivalency, such as combining multiple trapping centers to an equivalent trapping center. Regularization may be used in the loss calculation, such as where just the electrode signals are used, to deal with the reduced data available as ground truth.

Abstract: A weakly supervised intracranial hemorrhage (ICH) detection workflow includes training a deep learning (DL) model including a coupled convolutional neural network and recurrent neural network on a large dataset of CT scans with expert-labeled slices indicating presence or absence of ICH. Transfer learning (TL) is used to further train the DL model using a second large dataset of CT scans with only scan labels extracted from radiology reports using natural language processing (NLP). The DL model weights each slice of the scan against the final ICH diagnosis using an attention-based bi-directional long-short term memory network, where the attention weights represent slice-level ICH predictions. Model-generated heatmaps highlight significant regions of the CT scans that lead to the provided ICH predictions.

Abstract: A physics-based network model is trained to learn weights such as trapping, detrapping, and/or transport of holes and/or electrons, as well as voltage distribution on a voxel-by-voxel basis throughout a solid-state detector model. The physics-based network may be used to estimate material property variation throughout the voxels. Anode and cathode signals as well as the voltage distribution are relatively strong signals compared to the weaker electron and hole signals. The relatively weaker signals may be limited in range across voxels. In order to expand the range or magnify the effect, the loss function used in training the physics-based neural network may use a weighted combination where the weaker signals are weighted more heavily than stronger signals without substantially reducing the influence of the stronger signals. This improves the inference, resulting in improvement of the accuracy and range of the trained physics-based model.

Abstract: A system and a method perform frame interpolation for a compressed video bitstream. The system and the method may combine candidate pictures to generate an interpolated video picture inserted between two original video pictures. The system and the method may generate the candidate pictures from different motion fields. The candidate pictures may be generated partially or wholly from motion vectors extracted from the compressed video bitstream. The system and the method may reduce computation required for interpolation of video frames without a negative impact on visual quality of a video sequence.

Abstract: A method and apparatus for determining if a first video segment matches a second video segment is provided herein. Each video segment to be compared has an associated metric (HR), which is a function over time as the conditional entropy between frame fk and previous frame fk?1. A comparison of each video segment's HR vectors determines if the video segments match.

Abstract: A method and apparatus provides information for use in still image and video image processing, the information including scene and camera information and information obtained by sampling pixels or pixel regions during image formation. The information is referred to as meta-data. The meta-data regarding the camera and the scene is obtained by obtaining camera and sensor array parameters, generally prior to image acquisition. The meta-data obtained during the image formation obtained by sampling the pixels or pixel regions may include one or more masks marking regions of the image. The masks may identify blur in the image, under and/or overexposure in the image, and events occurring over the course of the image. Blur is detected by a sensing a change in pixel or pixel regions signal build up rate during the image acquisition. Under or over exposure is determined by pixels being below or above, respectively low and high thresholds.

Abstract: The present invention provides a method (200, 400) and device (500) for, within a variable or fixed block size video compression scheme, providing optimal bit allocation among at least three critical types of data: segmentation, motion vectors and prediction error, or DFD. Since the amount of information represented by one bit for a particular type of data is not equivalent to the information represented by one bit for some other data type, this consideration is taken into account to efficiently encode the video sequence. Thus, a computationally efficient method is provided for optimally encoding a given frame of a video sequence wherein, for a given bit budget the proposed encoding scheme leads to the smallest possible distortion and vice versa, for a given distortion, the proposed encoding scheme leads to the smallest possible rate.