Abstract: An unsupervised technique for training a deep learning based temporal noise reducer on unlabeled real-world data. The unsupervised technique can also be used to calibrate the free parameters of a TNR based on algorithmic principles. The training is based on actual real-world video (which may include noise), and not based on video containing artificial or added noise. Using the unsupervised technique to train a TNR allows the TNR to be tailored to the noise statistics of the use-case, resulting in the provision of high quality video with minimal resources. The TNR can be based on an uncalibrated TNR's output in time-reverse, as well as the uncalibrated TNR's output in time-forward. The frames used for both the time-forward output and the time-reversed output can be frames from the past. The TNR is calibrated to minimize the difference between its time-forward output and its time-reversed output.

Abstract: An example apparatus for video imaging includes a feature estimator to calculate a local value of a feature for averaging in a compressed set of features of a current frame. The apparatus also includes a validator to calculate a validity map comprising a weight for frame-wide averaging based on the compressed current frame. The apparatus further includes a vector generator to generate a state vector based on the local value of the feature and the weight. The apparatus further includes a relevance calculator to calculate a relevance to local processing for each coordinate in a restored state vector associated with a previous frame. The apparatus further includes a vector modulator to multiply the restored state vector by the relevance feature to generate a set of output features.

Abstract: An example apparatus for video frame segmentation includes a receiver to receive a current video frame to be segmented. The apparatus also includes a segmenting neural network to receive a previous mask including a segmentation mask corresponding to a previous frame and generate a segmentation mask for the current frame based on the previous mask and the video frame.

Abstract: A method, system, and article of highly efficient neural network video image processing uses temporal correlations.

Abstract: A method, system, and article is directed to real-time super-resolution image processing using neural networks.

Abstract: An example apparatus for video frame segmentation includes a receiver to receive a current video frame to be segmented. The apparatus also includes a segmenting neural network to receive a previous mask including a segmentation mask corresponding to a previous frame and generate a segmentation mask for the current frame based on the previous mask and the video frame.

Abstract: Techniques related to temporal noise reduction in captured video are discussed. Such techniques include performing motion estimation on a portion of a downsampled current frame performed during the downsampling of the current frame, replacing one or more of the resultant motion vectors based on confidence scores of the resultant motion vector, and blending the current frame and a temporally previous frame to generate a temporally filtered current frame. The temporally filtered current frame may be displayed to a user and/or encoded to generate a bitstream.

Abstract: An example apparatus for video imaging includes a feature estimator to calculate a local value of a feature for averaging in a compressed set of features of a current frame. The apparatus also includes a validator to calculate a validity map comprising a weight for frame-wide averaging based on the compressed current frame. The apparatus further includes a vector generator to generate a state vector based on the local value of the feature and the weight. The apparatus further includes a relevance calculator to calculate a relevance to local processing for each coordinate in a restored state vector associated with a previous frame. The apparatus further includes a vector modulator to multiply the restored state vector by the relevance feature to generate a set of output features.

Abstract: Techniques related to temporal noise reduction in captured video are discussed. Such techniques include performing motion estimation on a portion of a downsampled current frame performed during the downsampling of the current frame, replacing one or more of the resultant motion vectors based on confidence scores of the resultant motion vector, and blending the current frame and a temporally previous frame to generate a temporally filtered current frame. The temporally filtered current frame may be displayed to a user and/or encoded to generate a bitstream.