Microsoft Technology Licensing, LLC

Jingjing Fu

Yan Lu

Shipeng Li

Feng Wu

Bing Zeng

Dan Miao

Yao Zhai

Implementations of the subject matter described herein relate to object detection based on deep neural network. With a given input image, it is desired to determine a class and a boundary of one or more objects within the input image. Specifically, a plurality of channel groups is generated from a feature map of an image, the image including at least a region corresponding to a first grid. A target feature map is extracted from at least one of the plurality of channel groups associated with a cell of the first grid. Information related to an object within the region is determined based on the target feature map. The information related to the object may be a class and/or a boundary of the object.

Object detection based on a feature map of a convolutional neural network

Techniques for setting depth values for invalid measurement regions of depth images are described herein. A computing device may set the depth values based on evaluations of depth values of neighboring pixels and of corresponding pixels from time-adjacent depth images. Alternately or additionally, the computing device may utilize a texture image corresponding to the depth image to identify objects and may set depth values for pixels based on depth values of other pixels belonging to the same object. After setting the depth values, the computing device may normalize the depth values of the pixels. Further, the computing device may generate reduced representations of the depth images based on a depth reference model or a depth error model and may provide the reduced representations to an encoder.

Depth data processing and compression

In some examples, a layered encoding component and a layered decoding component provide for different ways to encode and decode, respectively, video streams transmitted between devices. For instance, in encoding a video stream, a layered encoding component may analyze the content of successive video frames and determine different types of encoding techniques to use for different ones of the video frames. Further, in some cases, some of the encoding techniques may be used on less than an entire video frame. In another example, in decoding a video stream, a layered decoding component may receive video frames encoded with different types of encoding. The layered decoding component may decode the differently encoded video frames and combine them to reconstruct a video stream.

Layered video encoding and decoding

The techniques and arrangements described herein provide for layered compression of depth image data. In some examples, an encoder may partition depth image data into a most significant bit (MSB) layer and a least significant bit (LSB) layer. The encoder may quantize the MSB layer and generate quantization difference data based at least in part on the quantization of the MSB layer. The encoder may apply the quantization difference data to the LSB layer to generate an adjusted LSB layer.

Representation and compression of depth data

Aspects of the subject matter described herein relate to image restoration for compressed images. In aspects, image restoration is accomplished by recovering spectral information from data corresponding to a compressed image. The spectral information is recovered using an algorithm to search through a solution space of possible solutions while constraints are imposed on the solution space to trim undesirable solutions from the space. An algorithm described herein may be iteratively applied to improve the quality of the recovered image.

Growing community of inventors

Beijing, China

Jingjing Fu