NEURAL NETWORK MODEL TRAINING METHOD, DEVICE AND STORAGE MEDIUM FOR IMAGE PROCESSING

摘要

Disclosed are a neural network model training method, device and storage medium for image processing, the method comprising: acquiring multiple temporally adjacent video frames (S202); respectively subjecting the multiple video frames to processing by a neural network model so as to enable the neural network model to output a corresponding intermediate image (S204); acquiring, from the multiple temporally adjacent video frames, optical flow information from which a video frame of an earlier time position is changed to a video frame of a later time position (S206); acquiring an image of the intermediate image corresponding to the video frame of an earlier time position changed according to the optical flow information (S208); acquiring a time loss between an intermediate image corresponding to the video frame of a later time position and a changed image (S210); acquiring a feature loss between intermediate images corresponding to the multiple temporally adjacent video frames and a target feature image (S212); and adjusting the neural network model according to the time loss and the feature loss, returning to the step of acquiring multiple temporally adjacent video frames and continuing with training until the neural network model satisfies a training ending condition (S214).