No abstract

No abstract

Figure 1. Gaussian noise image (left), our denoised image (middle) and ground truth photorealistic image (right).

Neural networks are often quantized to use reduced-precision arithmetic, as it greatly improves their storage and computational costs. This approach is commonly used in image classification and natural language processing applications. However, using a quantized network for the reconstruction of HDR images can lead to a significant loss in image quality. In this paper, we introduce QW-Net , a neural network for image reconstruction, in which close to 95% of the computations can be implemented with 4-bit integers. This is achieved using a combination of two U-shaped networks that are specialized for different tasks, a feature extraction network based on the U-Net architecture, coupled to a filtering network that reconstructs the output image. The feature extraction network has more computational complexity but is more resilient to quantization errors. The filtering network, on the other hand, has significantly fewer computations but requires higher precision. Our network recurrently warps and accumulates previous frames using motion vectors, producing temporally stable results with significantly better quality than TAA, a widely used technique in current games.

One of the significant challenges of care transitions in Intensive Care Units (ICUs) is the lack of effective support tools for outgoing clinicians to find, filter, organize, and annotate information that can be effectively handed off to the incoming team. We present a large display interactive multivariate visual approach, aimed towards supporting clinicians during the transition of care. We first provide a characterization of the problem domain in terms of data and tasks, based on an observation session at the University of Illinois Hospital, and on interviews with several biomedical researchers and ICU clinicians. Informed by this experience, we design a scalable, interactive visual approach that supports both overview and detail views of ICU patient data, as well as anomaly detection, comparison, and annotation of the data. We demonstrate a large-display implementation of the visualization on an existing anonymized ICU dataset. Feedback from domain experts indicates this approach successfully meets the requirements of effective care transitions.