Processing image data using deep neural networks is critical to many systems that desire to understand objects and their environment using camera sensors. Image scaling is a fundamental processing task required when managing image data. Although it is possible to scale image data using standard computer or graphics processors it would be highly advantageous in terms of throughput, latency and power consumption to manage image scaling using dedicated neural network hardware. The inventions contained herein provides methods to use existing neural network hardware to preform image scaling functions. Further, the inventions contained herein describe additional circuitry that can be added to neural network hardware to further enhance image scaling capabilities and efficiencies.

Systems and methods for generating interpolated images are disclosed. In examples, image features are extracted from a first image and a second image; such image features may be warped using first and second plurality of parameters. A first candidate intermediate frame may be generated based on the warped first features and the warped second features. Multi-scale features associated with the image features extracted from the first image and the second image may be obtained and warped using the first and second plurality of parameters. A second candidate intermediate frame may be generated based on the warped first multi-scale features and the warped second multi-scale features. By blending the first candidate intermedia frame with the second candidate intermediate frame, an interpolated image may be generated.

A method for developing an enhancement model for low-quality visual data, the method comprising the steps of receiving one or more sections of higher-quality visual data; and training a hierarchical algorithm. The hierarchical algorithm is operable to increase the quality of one or more sections of lower-quality visual data so as to substantially reproduce the one or more sections of higher-quality visual data. The hierarchical algorithm is then outputted.

A system and method for processing an input video while maintaining temporal consistency across video frames is provided. The method includes converting the input video from a first frame rate to a second frame rate, wherein the second frame rate is a faster frame rate than the first frame rate; generating processed frames of the input video at the second frame rate; and aggregating the processed frames using temporal sliding window aggregation to yield a processed output video at a third frame rate.

Methods and apparatuses for video transcoding based on spatial or temporal importance include: in response to receiving an encoded video bitstream, decoding a picture from the encoded video bitstream; determining a first level of spatial importance for a first region of a background of the picture based on an image segmentation technique; applying to the first region a first resolution-enhancement technique associated with the first level of spatial importance for increasing resolution of the first region by a scaling factor, wherein the first resolution-enhancement technique is selected from a set of resolution-enhancement techniques having different computational complexity levels; and encoding the first region using a video coding standard.

An image-recognition method is provided. The method includes the following steps: receiving structured data, wherein the structured data includes training-set data and testing-set data, and the structured data includes a plurality of groups, and each group includes one or more types, and each type includes a plurality of check-point images; training an artificial-intelligence (AI) model using the training-set data; inputting the testing-set data into the AI model to obtain a model evaluation of the AI model; and determining one or more first types with a lower overall recognition rate or a lower confidence level in the structured data, and deletes or corrects the check-point images in the one or more first types to update the structured data.

Examples relate to image processing, and performing fixed length format cell compression on a cell of a planar colour image based on an amount of white colour of the cell, the cell comprising a plurality of pixels, to obtain a compressed cell having four or fewer colour levels; and performing variable length format cell compression on the compressed cell to obtain a coded compressed cell.

This application discloses an image processing method, including: obtaining an image; performing feature extraction on the image to obtain at least one first feature map, where the at least one first feature map includes N first feature values, and N is a positive integer; obtaining a target compression bit rate, where the target compression bit rate corresponds to M target gain values, each target gain value corresponds to one first feature value, and M is a positive integer less than or equal to N; respectively processing corresponding first feature values based on the M target gain values to obtain M second feature values; and performing quantization and entropy encoding on at least one processed first feature map to obtain encoded data, where the at least one processed first feature map includes the M second feature values. Thus, compression bit rate control can be implemented in a same compression model.

An artificial intelligence (AI)-based frame interpolation method includes obtaining, from among consecutive frames of an image, feature maps for a first frame at a plurality of levels and feature maps for a second frame at the plurality of levels, obtaining, via a flow estimation neural network, a first optical flow from a first feature map at a certain level to a second feature map at the certain level and a second optical flow from the second feature map at the certain level to the first feature map at the certain level, and obtaining a forward-warped first feature map by forward-warping the first feature map using the first optical flow and a forward-warped second feature map by forward-warping the second feature map using the second optical flow.

A method for correcting an image divides an output image into a grid with vertical sections of width smaller than the image width but wide enough to allow efficient bursts when writing distortion corrected line sections into memory. A distortion correction engine includes a relatively small amount of memory for an input image buffer but without requiring unduly complex control. The input image buffer accommodates enough lines of an input image to cover the distortion of a single most vertically distorted line section of the input image. The memory required for the input image buffer can be significantly less than would be required to store all the lines of a distorted input image spanning a maximal distortion of a complete line within the input image.