Beijing Xiaomi Pinecone Electronics Co., Ltd.

An image processing method and apparatus, a model training method and apparatus, and a storage medium. The image processing method includes: acquiring a to-be-processed image; and inputting the to-be-processed image into a chromatic aberration-removing model to obtain a target image output by the chromatic aberration-removing model, where the chromatic aberration-removing model is obtained by a training data set through training, the training data set includes multiple groups of sample data, and each group of sample data includes a chromatic-aberration-free image and a chromatic-aberration image corresponding to the chromatic-aberration-free image.

A method for processing an image includes: creating M first image sets based on M image frames to be processed; determining a fusion weight for each sub-image located in each layer in each first image set based on brightness information of each image frame; fusing sub-images located in each same layer in the M first image sets based on the fusion weight, to obtain N fused sub-images arranged in layers based on the preset order; adjusting a brightness of at least one fused sub-image of the N fused sub-images based on a preset brightness adjustment parameter; and obtaining a target image based on the fused sub-image after adjustment.

An image processing method, a device and a non-transitory computer-readable storage medium thereof are provided. The method includes that a first image of a target object, a second image of the target object, and a third image of the target object are acquired. Exposure of the first image is less than exposure of the second image. The exposure of the second image is less than exposure of the third image. Further, brightness of the first image is increased to acquire a fourth image. Moreover, ghost elimination is performed on the second image and the fourth image is taken as a first reference frame to acquire a second reference frame. Furthermore, ghost elimination is performed on the third image based on the second reference frame to acquire a fifth image.

A method for optimizing a deep learning operator, includes: calling a method of reading an image object to read target data from an L1 cache of an image processor to the processor in response to detecting the target data in the L1 cache, performing a secondary quantization operation on the target data in the processor to obtain an operation result and writing the operation result into a main memory of the image processor. The target data is fixed-point data obtained after performing a quantization operation on data to be quantized in advance and the data to be quantized is one of the following: float-point data of an initial network layer of the neural network model and fixed-point data outputted from a network layer previous to the current network layer.

A method for video clip extraction includes: obtaining a video, and sampling the video to obtain N video frames, wherein N is a positive integer; inputting the N video frames to a pre-trained frame feature extraction model to obtain a feature vector of each video frame in N video frames; determining scores of the N video frames based on a pre-trained scoring model; and extracting target video clips from the video based on the scores of the N video frames.

A method for extracting a video clip includes obtaining a video and splitting the video into multiple clips. The multiple clips are input into a pre-trained scoring model, to obtain a score of each of the multiple clips. The scoring model is obtained by training based on data pairs including first clips and second clips, the data pairs including first clips and second clips are obtained based on labeled clips labeled with target attributes, and the target attributes include attributes that characterize clips as target clips or non-target clips. A target clip is extracted from the multiple clips based on the score of each of the multiple clips. An apparatus for extracting a video clip and a computer readable storage medium are also disclosed.

A method, apparatus, and a non-transitory computer-readable storage medium for processing a video are provided. A terminal determines a subject region of a video frame in a video and a background region. A target object is located in the subject region. The background region is a region of the video frame other than the subject region. The terminal overlays the subject region in at least one of a first video frame having the target object on at least one of a second video frame having the target object and generates a special effect frame including at least two subject regions in each of which the target object is located.

A training method for an image denoising model that can include collecting multiple sample image groups through a shooting device, each sample image group including multiple frames of sample images with a same photographic sensitivity and sample images in different sample image groups having different photographic sensitivities. The method can further include acquiring a photographic sensitivity of each sample image group, determining a noise characterization image corresponding to each sample image group based on the photographic sensitivity, determining a training input image group and a target image associated with each sample image group, each training input image group including all or part of sample images in a corresponding sample image group and a corresponding noise characterization image, constructing multiple training pairs each including a training input image group and a target image, and training the image denoising model based on the multiple training pairs until the image denoising model converges.

An image processing method and apparatus, a model training method and apparatus, and a storage medium. The image processing method includes: acquiring a to-be-processed image; and inputting the to-be-processed image into a chromatic aberration-removing model to obtain a target image output by the chromatic aberration-removing model, where the chromatic aberration-removing model is obtained by a training data set through training, the training data set includes multiple groups of sample data, and each group of sample data includes a chromatic-aberration-free image and a chromatic-aberration image corresponding to the chromatic-aberration-free image.

At least two images are acquired. At least two crop images are acquired by cropping the at least two images for face-containing images. Triangular patch deformation is performed on two neighbour images, generating a first triangular patch deformation frame image sequence and a second triangular patch deformation frame image sequence. Similarity transformation is performed on each image sequence of the first triangular patch deformation frame image sequence, acquiring a first transform frame image sequence. Similarity transformation is performed on each image sequence of the second triangular patch deformation frame image sequence, acquiring a second transform frame image sequence. The first and the second transform frame image sequences are fused, acquiring a video frame sequence corresponding to the two neighbour images. Of the at least two images, a video frame sequence generated by all neighbour images is coded, acquiring a destined video.