Transition frame prediction model training method and transition frame prediction method

Abstract

The invention provides a transition frame prediction model training method and a transition frame prediction method, and the training method comprises the steps of obtaining target joint data corresponding to a plurality of continuous motion frames, and the plurality of continuous motion frames comprise a key frame and an intermediate frame; masking the target joint data of the middle frame to obtain joint mask data of the middle frame; taking the joint mask data of the middle frame and the target joint data of the key frame as input of a self-attention neural network model, wherein the self-attention neural network model is constructed according to an affiliation relationship between joints; and adjusting the learnable parameters of the self-attention neural network model, so that the difference between the joint prediction result of the intermediate frame output by the self-attention neural network model and the target joint data of the intermediate frame is smaller than a threshold value, and a transition frame prediction model is obtained. The obtained transition frame prediction model fully considers the relation between the joints, the transition frame predicted by using the model conforms to the motion law of a human body, and the prediction accuracy of the transition frame is high.

Description

technical field [0001] The present application relates to the technical field of image processing, and in particular to a training method and device for a transition frame prediction model, a transition frame prediction method and device, electronic equipment, and a computer-readable storage medium. Background technique [0002] Character animation is a very important part of the game. Rich character animation can not only enrich the game content, but also improve the player's sense of substitution and gameplay. Generally speaking, a large-scale 3D game contains a huge number of character actions, and it takes a long time for a team of dozens of people to complete all the productions. [0003] At present, hand-made character animation is generally divided into three steps: setting key frames, that is, placing a specific action for the character at a specific position on the timeline, the denser the action, the smoother the generated character animation; action interpolation,...

AI Technical Summary

Problems solved by technology

[0004] Linear interpolation is only suitable for dense keyframes, but this will greatly increase the workload of art producers. If the number of keyframes is reduced, the accuracy of transition frame prediction will be reduced, which will make the movements of characters in the animation lack coherence

Images