Visual Data Coding via Dynamic Neural Network Module Switching

The patent implements dynamic module enabling/disabling mechanisms where neural network modules can be selectively activated or deactivated based on coding scenarios. The syntax elements in the bitstream control whether specific neural network modules (such as autoregressive models) are enabled, allowing the coding system to adapt its architecture dynamically rather than being fixed, thus resolving the contradiction between adaptability and structural simplicity.

The patent applies partial action by selectively enabling neural network modules only when needed rather than always activating them. Through syntax elements in the bitstream, the system determines whether to enable specific modules (such as autoregressive models), applying neural network processing only in scenarios where it provides necessary coding effectiveness, thus balancing coding performance with computational efficiency.

The patent segments the neural network coding system into modular components that can be independently enabled or disabled. By dividing the system into separate functional modules (such as autoregressive model modules, entropy coding modules, etc.), each with specific functions, the system can selectively activate only the necessary segments for a given coding task, thereby enhancing coding efficiency through specialized modules while managing overall system complexity through selective activation.