Abstract: Embodiments of standards-compliant compression of high-bit-depth visual data are disclosed herein. In one example, visual data is received in a first format, where the first format corresponds to a first color space having a first bit depth, and where the visual data is represented in the first color space. The visual data is rearranged from the first format into a second format, where the second format corresponds to a second color space having a second bit depth, and where the rearranged visual data in the second format remains represented in the first color space. The rearranged visual data in the second format is then encoded using a codec for the second color space.

Abstract: Devices and techniques related to implementing patch based video coding for machines are discussed. Such patch based video coding includes detecting regions of interest in a frame of video, extracting the detected regions of interest to one or more atlases absent the frame at a resolution not less than the resolution of the regions of interest, and encoding the one or more atlases to a bitstream.

Abstract: A Media Analytics Co-optimizer (MAC) engine that utilizes available motion and scene information to increase the activation sparsity in artificial intelligence (AI) visual media applications. In an example, the MAC engine receives video frames and associated video characteristics determined by a video decoder and reformats the video frames by applying a threshold level of motion to the video frames and zeroing out areas that fall below the threshold level of motion. In some examples, the MAC engine further receives scene information from an optical flow engine or event processing engine and reformats further based thereon. The reformatted video frames are consumed by the first stage of AI inference.

Abstract: In one embodiment, an electronic device includes a video-in interface, a video-out interface, memory circuitry, and processing circuitry. A first video stream with uncompressed frames is received via the video-in interface. The first video stream is compressed and then stored in a video buffer on the memory circuitry. For example, the uncompressed frames are individually compressed and stored in the video buffer. A second video stream with encoded frames is decoded and then played on a display device. For example, the encoded frames are decoded and then displayed on the display device via the video-out interface. The first video stream is then decompressed and played on the display device. For example, the compressed frames in the video buffer are individually decompressed and then displayed on the display device via the video-out interface.

Abstract: Disclosed examples include video frame segmenter circuitry to generate segmentation data of first video frame pixel data, the segmentation data including metadata corresponding to a foreground region and a background region, the foreground region corresponding to the first video frame pixel data. The disclosed examples also include video encoder circuitry to generate a first foreground bounding region and a first background bounding region based on the segmentation data, determine a first virtual tile of the first video frame pixel data, the first virtual tile located in the first foreground bounding region, encode the first virtual tile into a video data bitstream without encoding the first background bounding region, and transmit the video data bitstream via a network.

Abstract: In one embodiment, a compute device includes interface circuitry and processing circuitry. The processing circuitry receives, via the interface circuitry, a current frame of a video stream to be encoded. The processing circuitry then determines whether a scene change occurs at the current frame. If a scene change occurs at the current frame, the processing circuitry detects the scene in the current frame by performing pixel segmentation on the current frame. If a scene change does not occur at the current frame, the processing circuitry detects the scene in the current frame by performing motion estimation on the current frame relative to a previous frame in which the scene was detected. Based on the scene detected in the current frame, the processing circuitry then generates one or more encoding parameters and provides those parameters to a video encoder to encode the current frame.

Abstract: Embodiments are generally directed to video analytics encoding for improved efficiency of video processing and compression. An embodiment of an apparatus includes a memory to store data, including data for video streaming, and a video processing mechanism, wherein the video processing mechanism is to analyze video data and generate video analytics, generate metadata representing the video analytics and insert the generated video analytics metadata into a message, and transmit the video data and the metadata to a succeeding apparatus or system in a video analytics pipeline, the video data being compressed video data.

Abstract: Techniques for region-of-interest video encoding are disclosed. A compute node can determine a weighted average noise parameter for an encoded frame with a region of interest by weighting noise in the region of interest differently from noise outside the region of interest. Such a weighted average noise parameter can be used to encode subsequent frames of the video to improve the overall quality of the encoding. Other applications of the weighted average noise parameter are disclosed.

Abstract: Techniques for memory-efficient video encoding and decoding are disclosed. In the illustrative embodiment, a video encoder of a compute node uses a lossy compression algorithm to store a reference image of a video stream. The lossily compressed reference frame is then used to encode subsequent image of a video stream. When the corresponding decoder receives the reference image, it applies the same lossy compression algorithm to store the reference image. The lossily compressed reference frame is then used to decode subsequent images. Drift between the encoder and decoder caused by compression of the reference frame can be avoided by using the same lossy compression algorithm at both the video encoder and the video decoder.

Abstract: Methods, apparatus, systems, and articles of manufacture to compress weights of an artificial intelligence model are disclosed. An example apparatus includes a channel manipulator to manipulate weights of a channel of a trained model to generate a manipulated channel; a comparator to determine a similarity between (a) at least one of the channel or the manipulated channel and (b) a reference channel; and a data packet generator to, when the similarity satisfies a similarity threshold, generate a compressed data packet based on a difference between (a) the at least one of the channel or the manipulated channel and (b) the reference channel.

Abstract: Embodiments are generally directed to video analytics encoding for improved efficiency of video processing and compression. An embodiment of an apparatus includes a memory to store data, including data for video streaming, and a video processing mechanism, wherein the video processing mechanism is to analyze video data and generate video analytics, generate metadata representing the video analytics and insert the generated video analytics metadata into a message, and transmit the video data and the metadata to a succeeding apparatus or system in a video analytics pipeline, the video data being compressed video data.

Abstract: Embodiments are generally directed to video analytics encoding for improved efficiency of video processing and compression. An embodiment of an apparatus includes a memory to store data, including data for video streaming, and a video processing mechanism, wherein the video processing mechanism is to analyze video data and generate video analytics, generate metadata representing the video analytics and insert the generated video analytics metadata into a message, and transmit the video data and the metadata to a succeeding apparatus or system in a video analytics pipeline, the video data being compressed video data.

Abstract: Methods and apparatus are described for vehicle-to-vehicle communication. Embodiments receive a data request from a remote vehicle for data not available to the remote vehicle. Visual data from a host vehicle responsive to the data request that includes the data not available to the remote vehicle is determined. The visual data is analyzed for objects to create metadata associated with the visual data. The visual data and the metadata are provided to the remote vehicle in response to the data request.

Abstract: Methods and apparatus relating to deep learning solutions for safe, legal, and/or efficient autonomous driving are described. In an embodiment, first logic determines a geographic location of a vehicle, a weather condition at the geographic location, and a maneuver for the vehicle based at least in part on sensor data and a target location. Memory stores data corresponding to the geographic location, the weather condition, and the maneuver. The first logic causes one or more motion planning logic to actuate or control movement of the vehicle based on the stored data. Other embodiments are also disclosed and claimed.