Sony Broadcast & Professional Research Laboratories

About: Sony Broadcast & Professional Research Laboratories is a company organization based out in Taipei, Taiwan. It is known for research contribution in the topics: Signal & Layer (electronics). The organization has 38708 authors who have published 63864 publications receiving 865637 citations.

LF-Net: Learning Local Features from Images

Abstract: We present a novel deep architecture and a training strategy to learn a local feature pipeline from scratch, using collections of images without the need for human supervision. To do so we exploit depth and relative camera pose cues to create a virtual target that the network should achieve on one image, provided the outputs of the network for the other image. While this process is inherently non-differentiable, we show that we can optimize the network in a two-branch setup by confining it to one branch, while preserving differentiability in the other. We train our method on both indoor and outdoor datasets, with depth data from 3D sensors for the former, and depth estimates from an off-the-shelf Structure-from-Motion solution for the latter. Our models outperform the state of the art on sparse feature matching on both datasets, while running at 60+ fps for QVGA images.

Method and system for remote access of personal music

Abstract: In one aspect of the invention, a method is provided for storing audio files. The method includes: (a) receiving electronic files at a central location from a first device, those electronic files representing audio signals; (b) associating the audio files with identification information; (c) storing the audio files at the central location on at least a portion of a storage media, that portion being uniquely associated with the identification information; (d) receiving the identification information from a second device; and (e) transmitting the audio files to the second device upon receipt of the identification information.

Method, system and devices for wireless data storage on a server and data retrieval

Abstract: A method and system for remotely storing data on a server through a wireless connection instead of storing data locally in a consumer device, as well as devices for use with the method and system. More particularly, a video camera, still camera, laptop computer, or other consumer device which normally stores data in local memory such as film, disk, random access memory, memory sticks, or other forms of storage would transmit the data to a remote server through a wireless connection. The data would be saved on the remote server for subsequent retrieval through, for example, the Internet or a wireless connection to the server. In addition, data not originating from the user device could be downloaded to the consumer device. The data to be retrieved can be specified by the user, or sent to the user according to a user profile stored on the server.

Content abstraction layer for use in home network applications

Abstract: A network architecture for a network of electronic devices includes a device layer having a plurality of electronic devices interconnected using a network backbone, wherein the plurality of electronic devices each operate using a device native communication protocol. The architecture also includes a device abstraction layer (DAL) which communicates with each of the devices using the device native communication protocols and also presents a unified communication interface to a content abstraction program interface. The content abstraction program interface communicates with the device layer through the unified communication interface of the DAL and includes a set of content services for controlling content on the network. In one embodiment the content abstraction program interface includes a content location system (CLS), a content change notification system (CCNS), and a content engagement system (CES).

Improving music source separation based on deep neural networks through data augmentation and network blending

Abstract: This paper deals with the separation of music into individual instrument tracks which is known to be a challenging problem. We describe two different deep neural network architectures for this task, a feed-forward and a recurrent one, and show that each of them yields themselves state-of-the art results on the SiSEC DSD100 dataset. For the recurrent network, we use data augmentation during training and show that even simple separation networks are prone to overfitting if no data augmentation is used. Furthermore, we propose a blending of both neural network systems where we linearly combine their raw outputs and then perform a multi-channel Wiener filter post-processing. This blending scheme yields the best results that have been reported to-date on the SiSEC DSD100 dataset.