DeepWave: A Recurrent Neural-Network for Real-Time Acoustic Imaging

Authors: Matthieu SIMEONI, Sepand Kashani, Paul Hurley, Martin Vetterli

NeurIPS 2019 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Our real-data experiments show Deep Wave has similar computational speed to the state-of-the-art delay-and-sum imager with vastly superior resolution. While developed primarily for acoustic cameras, Deep Wave could easily be adapted to neighbouring signal processing fields, such as radio astronomy, radar and sonar.
Researcher Affiliation Collaboration Matthieu Simeoni IBM Zurich Research Laboratory meo@zurich.ibm.com Sepand Kashani École Polytechnique Fédérale de Lausanne (EPFL) sepand.kashani@epfl.ch Paul Hurley Western Sydney University paul.hurley@westernsydney.edu.au Martin Vetterli École Polytechnique Fédérale de Lausanne (EPFL) martin.vetterli@epfl.ch
Pseudocode Yes Algorithm 1 Deep Wave forward propagation
Open Source Code Yes Deep Wave implementation can be found on https://github.com/imagingofthings/Deep Wave.
Open Datasets Yes Finally we express our gratitude towards Robin Scheibler and Hanjie Pan for their openly-accessible real-world datasets [36, 43].
Dataset Splits Yes Deep Wave is trained by splitting the data points into a training and validation set (respectively 80% and 20% in size).
Hardware Specification No The paper mentions 'a general-purpose CPU' and 'a standard computing platform' but does not provide specific models of CPUs, GPUs, or other hardware components used for running the experiments.
Software Dependencies No The paper does not explicitly state specific software dependencies with version numbers (e.g., Python, PyTorch, TensorFlow versions or library versions).
Experiment Setup Yes For each frequency band, we chose an architecture with 5 layers. Optimisation of (9) is carried out by stochastic gradient descent (SGD) with momentum acceleration [51]. Finally, we substitute the Re Lu activation function by a scaled rectified tanh to avoid the exploding gradient problem [39].