Video frame interpolation method and device, computer readable storage medium

A video frame interpolation device, comprising a processor and a memory, wherein the memory is configured to store program instructions, and the processor is configured to call the program instructions to perform a method comprising:

• inputting at least two image frames into a video frame interpolation model to obtain at least one frame-interpolation image frame, wherein the video frame interpolation model is obtained by training an initial model, and the training the initial model comprises:
• obtaining the initial model and sample images, and the training the initial model using a first loss to obtain a reference model;
• obtaining a first training model comprising three identical reference models, selecting two first target sample image frames from the sample images according to a preset rule, and inputting the two first target sample image frames into a first reference model of the first training model to obtain a first frame-interpolation result;
• selecting two second target sample image frames from the sample images according to the preset rule, and inputting the two second target sample image frames into a second reference model of the first training model to obtain a second frame-interpolation result;
• selecting one frame-interpolation image from the first frame-interpolation result according to the preset rule and selecting one frame-interpolation image from the second frame-interpolation result according to the preset rule, wherein the frame-interpolation images selected are used as two third target sample image frames that are input into a third reference model of the first training model for calculation to obtain a third frame-interpolation result;
• adjusting parameters of the first training model based on total loss of the first training model, wherein the total loss of the first training model is obtained based on the first frame-interpolation result, the second frame-interpolation result, and the third frame-interpolation result and the sample images, and the parameters are shared by each reference model of the first training model;
• and
• using a parameter model obtained via a predetermined number of iterations as the video frame interpolation model.

A non-transitory computer-readable storage medium storing a computer program, wherein the computer program, when executed by a processor, implements a video frame interpolation method, comprising:

• inputting at least two image frames into a video frame interpolation model to obtain at least one frame-interpolation image frame, wherein the video frame interpolation model is obtained by training an initial model and the training the initial model comprises:
• obtaining the initial model and sample images, and the training the initial model by using a first loss to obtain a reference model;
• obtaining a first training model comprising three identical reference models, selecting two first target sample image frames from the sample images according to a preset rule, and inputting the first target sample image frames into a first reference model of the first training model to obtain a first frame-interpolation result;
• selecting two second target sample image frames from the sample images according to the preset rule, and inputting the second target sample image frames into a second reference model of the first training model to obtain a second frame-interpolation result;
• selecting one frame-interpolation image from the first frame-interpolation result according to the preset rule and selecting one frame-interpolation image from the second frame-interpolation result according to the preset rule, wherein the frame-interpolation images selected are used as two third target sample image frames that are input into a third reference model of the first training model for calculation to obtain a third frame-interpolation result;
• adjusting parameters of the first training model based on total loss of the first training model, wherein the total loss of the first training model is obtained based on the first frame-interpolation result, the second frame-interpolation result, and the third frame-interpolation result and the sample images, and the parameters are shared by each reference model of the first training model;
• and
• using a parameter model obtained via a predetermined number of iterations as the video frame interpolation model.