A method of video decoding performed in a video decoder includes receiving a coded video bitstream including a current block that is divided into a plurality of sub-blocks. Sub-block based affine motion compensation is performed on the current block to generate a sub-block prediction for each pixel in each sub-block of the current block. One or more spatial gradients are determined for each sub-block prediction. For each sub-block prediction, prediction refinement with optical flow process is performed using the respective determined one or more spatial gradients and at least one constraint included in the coded video bitstream. For each sub-block prediction, an output of the respective prediction refinement is added to the respective sub-block prediction to generate a final prediction for each pixel in each sub-block of the current block.

The present disclosure provides a video image decoding method in which a size of an affine motion compensation image sub-block in an affine image block is determined based on a motion vector difference, motion vector precision, a distance between control points in the affine image block, and a size of the affine image block, where the size includes a length in a horizontal direction and a length in a vertical direction, so that a length of the affine image block in a horizontal/vertical direction is an integer multiple of the length of the affine motion compensation image sub-block in the horizontal/vertical direction.

A method of video processing includes deriving, for a current block, a set of control point motion vectors; determining a motion prediction model for the current block based on the set of CPMVs; and performing a conversion between the current block and a bitstream representation of the current block using the motion prediction model, the motion prediction model being a 8-parameter affine prediction model.

Systems, methods, and instrumentalities for affine motion estimation for affine model-based video coding may be disclosed herein. A first motion vector (MV) set including one or more MVs may be derived for a first coding block. The MVs may be control point MVs (CPMVs) and the MVs may be derived by performing affine motion estimation (ME) associated with the first coding block. The first MV set may be added to a recently-estimated MV list. A head of the recently-estimated MV list may be set to the first MV set. The recently-estimated MV list may be empty or may contain one or more previously-added MV sets.

The present disclosure provides a video picture coding method, a video picture decoding method, a coding device, and a decoding device. The method includes: determining a distance between control points for an affine picture block; determining a motion vector difference for the affine picture block, motion vectors of the control points being used to determine the motion vector difference; and performing coding processing on the affine picture block over a size that includes a horizontal length and a vertical length, wherein the horizontal length and the vertical length vary based on the distance between the control points, the motion vector difference, and a motion vector precision.

The present disclosure provides a video picture coding method, a video picture decoding method, a coding device, and a decoding device. The method includes: determining a distance between control points for an affine picture block; determining a motion vector difference for the affine picture block, motion vectors of the control points being used to determine the motion vector difference; and performing coding processing on the affine picture block over a size that includes a horizontal length and a vertical length, wherein the horizontal length and the vertical length vary based on the distance between the control points, the motion vector difference, and a motion vector precision.

An image decoding method according to the present document includes obtaining motion prediction information for a current block from a bitstream, generating an affine MVP candidate list for the current block, deriving CPMVPs for CPs of the current block based on the affine MVP candidate list, deriving CPMVDs for the CPs of the current block based on the motion prediction information, deriving CPMVs for the CPs of the current block based on the CPMVPs and the CPMVDs, and deriving prediction samples for the current block based on the CPMVs.

Methods, systems and device for hash-based motion estimation in video coding are described. An exemplary method of video processing includes determining, for a conversion between a current block of a video and a bitstream representation of the video, motion information associated with the current block using a hash-based motion search, a size of the current block being M×N, M and N being positive integers and M being not equal to N, applying, based on the motion information and a video picture comprising the current block, a prediction for the current block, and performing, based on the prediction, the conversion.

Provided is an encoder including circuitry and memory. In operation, the circuitry: derives one or more first motion vectors based on at least one of one or more motion vectors of a first block encoded in an affine mode; derives one or more second motion vectors based on at least one of one or more motion vectors of a second block that has been encoded and is different from the first block; and determines, as motion vector predictors at control points of a current block to be encoded, motion vectors including at least one of the first motion vectors and at least one of the second motion vectors.

A moving image decoding method for generating prediction images 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 horizontally neighboring samples and vertically neighboring samples in the first and second prediction images 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 third 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 derived based on a pixel bit length of eight.