Neural network model training method and device, and time-lapse photography video generating method and device

An apparatus for generating a time-lapse photography video with a neural network model, the apparatus comprising:

• a memory storing instructions;
• and
• a processor in communication with the memory, wherein, when the processor executes the instructions, the processor is configured to cause the apparatus to:
• obtain a training sample, the training sample comprising a training video and an image set corresponding to the training video,
• obtain a basic time-lapse photography video by using a first generative adversarial network of a neural network model,
• obtain an optimized time-lapse photography video based on the basic time-lapse photography video by using a generator in a second generative adversarial network of the neural network model,
• obtain a discrimination result according to the optimized time-lapse photography video by using a discriminator in the second generative adversarial network,
• generate a loss of the second generative adversarial network according to the optimized time-lapse photography video, the basic time-lapse photography video, the training video, and the discrimination result, the loss comprising a ranking loss determined according to motion features respectively corresponding to the optimized time-lapse photography video, the basic time-lapse photography video, and the training video, and
• optimize a set of network parameters of the second generative adversarial network according to the loss of the second generative adversarial network, until a training ending condition is satisfied.

A non-transitory computer readable storage medium storing computer readable instructions, wherein, the computer readable instructions, when executed by a processor, are configured to cause the processor to perform:

• obtaining a training sample, the training sample comprising a training video and an image set corresponding to the training video;
• obtaining a basic time-lapse photography video by using a first generative adversarial network of a neural network model;
• obtaining an optimized time-lapse photography video based on the basic time-lapse photography video by using a generator in a second generative adversarial network of the neural network model;
• obtaining a discrimination result according to the optimized time-lapse photography video by using a discriminator in the second generative adversarial network;
• generating a loss of the second generative adversarial network according to the optimized time-lapse photography video, the basic time-lapse photography video, the training video, and the discrimination result, the loss comprising a ranking loss determined according to motion features respectively corresponding to the optimized time-lapse photography video, the basic time-lapse photography video, and the training video;
• and
• optimizing a set of network parameters of the second generative adversarial network according to the loss of the second generative adversarial network, until a training ending condition is satisfied.