Video Codec Connectors for Human and Machine Vision Outputs

The patent segments the decoding process into two distinct paths: a first decoding path optimized for human viewing quality and a second decoding path optimized for machine task accuracy. This segmentation allows each path to use different processing parameters and algorithms tailored to its specific purpose, resolving the contradiction between visual quality and machine task performance

The patent creates a universal decoding system that can serve multiple functions: it can output decoded video optimized for human viewing, optimized for machine tasks, or both simultaneously. The multi-functional decoder includes multiple output interfaces that can be selectively activated based on the intended use case, allowing the same compression system to satisfy both human visual quality requirements and machine task accuracy requirements

The patent implements dynamic parameter adjustment in the second decoding path, where processing parameters such as filtering strength, block sizes, and algorithm selection are dynamically adapted based on the specific machine task requirements and available bitrate. This dynamic adjustment allows the system to maintain task accuracy across varying compression levels by optimizing the decoding process for each specific scenario

The patent changes key decoding parameters in the second path compared to the first path, including using different filter types, block partitioning schemes, and reconstruction algorithms. These parameter changes are specifically designed to preserve features important for machine tasks even at low bitrates, while the first path uses parameters optimized for human visual perception