Video frame supplementing method based on neural network and training method of model thereof

The invention provides a video frame supplementing method based on a neural network and a training method of a model of the video frame supplementing method. The method comprises the following steps of after determining a current training reference frame based on a preset training set, inputting the training reference frame into a preset initial model; generating an initial feature map of a presetnumber of levels of the training reference frame through a feature extraction network; fusing the preset number of levels of initial feature maps into a fused feature map through a feature fusion network; inputting the fusion feature map into an output network, and outputting a training supplementary video frame between the first training frame and the second training frame; determining a loss value of the training supplementary video frame through a preset prediction loss function; and continuously inputting the next group of training reference frames into the initial model for training until the parameters in the i