3D Volumetric Data Encoding With Pose-Shape Residual Compression
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for encoding 3D volumetric data are inefficient due to its complexity and high capacity, particularly in virtual human models, where temporal and geometric structures vary significantly across frames, making it difficult to apply temporal correlation for compression.
Innovation Solution
A method that estimates pose and shape data from 3D volumetric data, generates a 3D estimation model by modifying a template model, and encodes residual data, involving quantization, 2D projection, and error correction from multiple viewpoints to reduce data complexity.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If 3D volumetric data is encoded using existing methods, then the data can be transmitted and stored, but the data amount remains very large and complex due to the high capacity and geometric variability of volumetric models
Solution Approach 1:
The encoding method segments the 3D volumetric data into multiple components: a base 3D model (from multi-view images), depth maps for each view, and normal maps. By dividing the complex volumetric data into these separable elements, each can be encoded and compressed independently, reducing the overall data amount while maintaining geometric accuracy
Solution Approach 2:
The patent introduces intermediate representations (depth maps and normal maps) as mediators between the multi-view images and the final 3D volumetric model. These intermediaries capture geometric information in a compressed form that can be efficiently transmitted and reconstructed, avoiding direct transmission of the full volumetric data
2Shape
If the mesh structure of each frame is highly variable, then the 3D volumetric data preserves vivid shapes and motions, but temporal correlation cannot be effectively applied for compression
Solution Approach 1:
The method performs preliminary actions by first generating a base 3D model from multi-view images before processing individual frames. This base model serves as a reference that captures the overall geometry, allowing subsequent frame differences to be encoded with higher compression efficiency while preserving geometric accuracy
Solution Approach 2:
The patent applies dynamics by using temporal prediction that adapts to frame-to-frame variations. The encoding process dynamically adjusts based on the variability between frames, applying compression techniques that respond to the actual geometric changes rather than using a static approach
Data Source
AI summary
Disclosed is a method for encoding 3D volumetric data to estimate pose and shape data from 3D volumetric data, generate a 3D estimation model by applying the estimated data to a 3D template model, and encode residual data of the 3D volumetric data together with the 3D estimation model, and the method includes: (a) receiving 3D volumetric data; (c) estimating a 3D pose from the 3D volumetric data; (d) estimating a 3D shape from the 3D volumetric data; (e) generating a 3D estimation model by modifying a predefined 3D template model; (f) calculating a residual of the 3D volumetric data with respect to the 3D estimation model; and (h) generating the pose estimation data, the shape estimation data and data on the residual as transmission data. Accordingly, an amount of data in the 3D volumetric model is significantly reduced and complexity is reduced.