Sorry, you need to enable JavaScript to visit this website.

Deep-learning based speech enhancement systems have offered tremendous gains, where the best performing approaches use long short-term memory (LSTM) recurrent neural networks (RNNs) to model temporal speech correlations. These models, however, do not consider the frequency-level correlations within a single time frame, as spectral dependencies along the frequency axis are often ignored. This results in inaccurate frequency responses that negatively affect perceptual quality and intelligibility. We propose a deep-learning approach that considers temporal and frequency-level dependencies.

After their triumph in various classification, recognition and segmentation problems, deep learning and convolutional networks are now making great strides in different inverse problems of imaging. Magnetic resonance image (MRI) reconstruction is an important imaging inverse problem, where deep learning methodologies are starting to make impact. In this work we will develop a new Convolutional Neural Network (CNN) based variant for MRI reconstruction. The developed algorithm is based on the recently proposed deep cascaded CNN (DC-CNN) structure.

Non-maximum suppression (NMS) is a post-processing step in almost every visual object detector. Its goal is to drastically prune the number of overlapping detected candidate regions-of-interest (ROIs) and replace them with a single, more spatially accurate detection. The default algorithm (Greedy NMS) is fairly simple and suffers from drawbacks, due to its need for manual tuning. Recently, NMS has been improved using deep neural networks that learn how to solve a spatial overlap-based detections rescoring task in a supervised manner, where only ROI coordinates are exploited as input.

In recent years, deep metric learning has achieved promising results in learning high dimensional semantic feature embeddings where the spatial relationships of the feature vectors match the visual similarities of the images. Similarity search for images is performed by determining the vectors with the smallest distances to a query vector. However, high retrieval quality does not depend on the actual distances of the feature vectors, but rather on the ranking order of the feature vectors from similar images.

We propose Shift R-CNN, a hybrid model for monocular 3D object detection, which combines deep learning with the power of geometry. We adapt a Faster R-CNN network for regressing initial 2D and 3D object properties and combine it with a least squares solution for the inverse 2D to 3D geometric mapping problem, using the camera projection matrix. The closed-form solution of the mathematical system, along with the initial output of the adapted Faster R-CNN are then passed through a final ShiftNet network that refines the result using our newly proposed Volume Displacement Loss.

Cloud segmentation is a vital task in applications that utilize satellite imagery. A common obstacle in using deep learning-based methods for this task is the insufficient number of images with their annotated ground truths. This work presents a content-aware unpaired image-to-image translation algorithm. It generates synthetic images with different land cover types from original images while preserving the locations and the intensity values of the cloud pixels. Therefore, no manual annotation of ground truth in these images is required.

Unsupervised object discovery in images involves uncovering recurring patterns that define objects and discriminates them against the background. This is more challenging than image clustering as the size and the location of the objects are not known: this adds additional degrees of freedom and increases the problem complexity. In this work, we propose StampNet, a novel autoencoding neural network that localizes shapes (objects) over a simple background in images and categorizes them simultaneously.

In-Loop filter is a key part in High Efficiency Video Coding(HEVC) which effectively removes the compression artifacts.Recently, many newly proposed methods combine residual learning and dense connection to construct a deeper network for better in-loop filtering performance. However,the long-term dependency between blocks is neglected, and information usually passes between blocks only after dimension compression.

A quantitative understanding of complex biological systems such as tissues requires reconstructing the structure of the different components of the system. Fluorescence microscopy provides the means to visualize simultaneously several tissue components. However, it can be time consuming and is limited by the number of fluorescent markers that can be used. In this study, we describe a toolbox of algorithms based on convolutional neural networks for the prediction of 3D tissue structures by learning features embedded within single-marker images.