A video decoder can be configured to determine that a current block in a current picture of the video data is coded in an affine prediction mode; determine one or more control-point motion vectors (CPMVs) for the current block; identify an initial prediction block for the current block in a reference picture using the one or more CPMVs; determine a current template for the current block in the current picture; and determine an initial reference template for the initial prediction block in the reference picture; and perform a motion vector refinement process to determine a modified prediction block based on a comparison of the current template to the initial reference template.

An example method of video processing includes partitioning, for a conversion between a current video block of a picture of a video and a bitstream of the video, the picture into one or more slices according to a requirement of bitstream conformance. The requirement of bitstream conformance specifies that a union of the one or more slices covers the whole picture, and a first syntax element is included in the bitstream to indicate whether the one or more slices have a rectangular shape. The method also includes performing the conversion at least based on the partitioning.

Apparatus and method for inter-frame prediction. For example, one embodiment of a method comprises extracting a first set of key points from a current frame and a second set of key points from a reference frame; projecting a key point p into the reference frame as p′ based on a motion vector associated a region in the current frame in which key point p is found; locating n key points in the reference frame which are closest to p′; and matching one of the n key points in the reference frame to key point p in the current frame based on descriptor distances between each of the n key points and p′.

Apparatus and method for inter-frame prediction. For example, one embodiment of a method comprises extracting a first set of key points from a current frame and a second set of key points from a reference frame; projecting a key point p into the reference frame as p′ based on a motion vector associated a region in the current frame in which key point p is found; locating n key points in the reference frame which are closest to p′; and matching one of the n key points in the reference frame to key point p in the current frame based on descriptor distances between each of the n key points and p′.

A method for decoding an image, according to the present invention, comprises the steps of: determining whether a coding block is divided into a first prediction unit and a second prediction unit; determining a partition type of the coding block when determining that the coding block is divided; deriving first motion information about a first prediction unit and second motion information about a second prediction unit within the coding block; and acquiring a prediction sample within the coding block on the basis of the first motion information and the second motion information.

The present invention relates to a method for encoding and decoding images, wherein, if bi-prediction is applied, a weight for a merge candidate is derived, thereby improving the image compression efficiency, and making it possible to efficiently configure a motion vector candidate. A merge candidate list of the current block is produced on the basis of inter-prediction information, and the candidate includes a configured affine merge candidate. If the configured affine merge candidate is produced on the basis of a combination of control points including CP0, weight index information regarding the configured affine merge candidate is produced on the basis of weight index information regarding CP0. If the configured affine merge candidate is produced on the basis of a combination of control points excluding CP0, weight index information regarding the configured affine merge candidate is derived on the basis of weight index information regarding CP1.

A video decoding method according to the present disclosure may include determining whether an affine motion model is applied to a current block or not, performing motion compensation for the current block according to whether the affine motion model is applied or not, determining a value of a first variable and a second variable representing whether a prediction block obtained by the motion compensation will be refined and determining a padding size of the prediction block.

Examples of video encoding methods and apparatus and video decoding methods and apparatus are described. An example method of processing video data includes determining, for a conversion between a video block of a video and a bitstream of the video, a gradient of a prediction vector at a sub-block level for the video block according to a rule, wherein the rule specifies to use a same gradient value is assigned for all samples within a sub-block of the video block; and performing the conversion based on the determining.

Techniques for implementing video processing techniques are described. In one example implementation, a method of video processing includes determining, for a conversion between a current block of a video coded using a combined inter and intra prediction (CIIP) coding technique and a bitstream representation of the video, an intra-prediction mode of the current block independently from an intra-prediction mode of a neighboring block. The CIIP coding technique uses an intermedia inter prediction value and an intermedia intra prediction value to derive a final prediction value of the current block. The method also includes performing the conversion based on the determining.

A prediction image generation method using two prediction images to generate a prediction image by a device is provided. First and second prediction image are generated. Bidirectional prediction gradient change prediction processing is performed by using a first shift value and difference values of the first and second prediction images respectively in horizontal and vertical directions to generate a first, second, third and fourth gradient images. Motion information is derived by using the first and second prediction images, the first, second, third and fourth gradient images, a second shift value, and a third shift value. Motion compensation correction value is derived by using the motion information and the first, second, third and the fourth gradient images. The prediction image is generated by using the first and second prediction images and the motion compensation correction value. The first, second and third shift values are respectively equal to 6, 4 and 1.