IP Library › Granted Patent US 12705800

Granted Patent B2

US 12705800 · App. 18/762,428 · Granted Aug 11, 2026

Quantization in mesh compression

Inventors: Jun Tian (Belle Mead, NJ); Xiaozhong Xu (Shenzhen, CN); Shan Liu (San Jose, CA)

Assignee: TENCENT AMERICA LLC

G06T9/001

View Patent ↗

Loading inventors, assignments & file history…

Monitor This Case

Get email alerts when status or documents change.

Order Certified Copies

Most orders are placed with the USPTO same day — all within 24 business hours.

Order via The Patent Place →

Pre-filled with this patent's details

Quick Facts

Patent No.: US 12705800
App. No.: 18/762,428
Granted: Aug 11, 2026
Kind: B2

External Links

USPTO Patent Center↗ Full Patent PDF↗

Abstract

In a method, a base mesh includes a subset of a plurality of vertices of a mesh. A position prediction of a current vertex of the base mesh is determined. A motion field prediction of the current vertex of the base mesh is determined. A position prediction residue for the position prediction of the current vertex is determined based on a first quantization step value. A motion field prediction residue for the motion field prediction of the current vertex is determined based on a second quantization step value, where the second quantization step value is dependent on the first quantization step value. A position of the current vertex of the base mesh is reconstructed based on the position prediction and the position prediction residue. A motion field of the current vertex of the base mesh is reconstructed based on the motion field prediction and the motion field prediction residue.

Claims (83)

1 . A method of mesh decoding, comprising:

receiving a bitstream that includes base mesh information of a base mesh, the base mesh including a subset of a plurality of vertices of a mesh in a current mesh frame;

determining (i) a position prediction of a current vertex of the base mesh, and (ii) a motion field prediction of the current vertex of the base mesh;

determining (i) a position prediction residue for the position prediction of the current vertex based on a first quantization step value, and (ii) a motion field prediction residue for the motion field prediction of the current vertex based on a second quantization step value, the second quantization step value being determined based on a function of the first quantization step value; and

reconstructing (i) a position of the current vertex of the base mesh based on the position prediction and the position prediction residue, and (ii) a motion field of the current vertex of the base mesh based on the motion field prediction and the motion field prediction residue.

2 . The method of claim 1 , wherein:

the determining the position prediction residue includes:

determining a quantized position prediction residue for the position prediction of the current vertex based on a first entropy coding, and

dequantizing the quantized position prediction residue for the position prediction of the current vertex to determine the position prediction residue based on the first quantization step value; and

the determining the motion field prediction residue includes:

determining a quantized motion field prediction residue for the motion field prediction of the motion field for the current vertex based on a second entropy coding, and

dequantizing the quantized motion field prediction residue for the motion field prediction of the current vertex to determine the motion field prediction residue based on the second quantization step value.

3 . The method of claim 2 , wherein the second quantization step value is equal to a multiple of the first quantization step value.

4 . The method of claim 2 , wherein the second quantization step value is equal to one of a linear function and a monotonically non-decreasing function of the first quantization step value.

5 . The method of claim 2 , wherein the second quantization step value is equal to a monotonically non-decreasing function of a quantization error associated with the position prediction of the current vertex, the quantization error being a difference between the quantized position prediction residue and an unquantized position prediction residue for the current vertex.

6 . The method of claim 2 , wherein the first quantization step value is defined by a positive dyadic rational of

m being a positive integer between 0 and 7, n being a non-negative integer between 1 and 256.

7 . The method of claim 2 , wherein:

the first quantization step value and the second quantization step value are received at a sequence level, and

the first quantization step value and the second quantization step value are equal when a syntax element in the bitstream indicates that the first quantization step value is equal to the second quantization step value.

8 . The method of claim 2 , wherein:

the first quantization step value and the second quantization step value are received at a frame level,

the second quantization step value is equal to a quantization step value of a reference frame of the current mesh frame when a syntax element in the bitstream is a first value, and

the second quantization step value is equal to the first quantization step value when the syntax element in the bitstream is a second value.

9 . The method of claim 1 , further comprising:

determining a two-dimensional (2D) texture coordinate prediction of a 2D texture coordinate of the current vertex of the base mesh;

determining a 2D texture coordinate prediction residue for a 2D texture coordinate prediction of the current vertex; and

reconstructing a 2D texture coordinate of the current vertex of the base mesh based on the 2D texture coordinate prediction and the 2D texture coordinate prediction residue.

10 . The method of claim 9 , wherein:

when the 2D texture coordinate of the current vertex is directly coded, the determining the 2D texture coordinate prediction residue includes:

determining a quantized 2D texture coordinate for the current vertex, the quantized 2D texture coordinate being coded by a fixed-length coding, and

dequantizing the quantized 2D texture coordinate for the current vertex to obtain the 2D texture coordinate, the quantized 2D texture coordinate being obtained by a rounding quantization, the rounding quantization being configured to convert the 2D texture coordinate into an integer.

11 . The method of claim 9 , wherein:

when the 2D texture coordinate of the current vertex is predicted by a stretch prediction algorithm, the determining the 2D texture coordinate prediction residue includes:

determining a quantized 2D texture coordinate prediction residue for the current vertex, the quantized 2D texture coordinate prediction residue being coded by a variable length coding, and

dequantizing the quantized 2D texture coordinate prediction residue for the current vertex to determine the 2D texture coordinate prediction residue, the quantized 2D texture coordinate prediction residue being obtained by a rounding quantization, the rounding quantization being configured to convert the 2D texture coordinate prediction residue into an integer.

12 . The method of claim 11 , wherein:

the rounding quantization is defined by a dyadic rational of

m being a positive integer between 0 and 7, n being a non-negative integer between 0 and 255, and

the quantized 2D texture coordinate prediction residue is obtained by rounding a multiple of the 2D texture coordinate prediction residue and an inverse of the dyadic rational.

13 . A method of mesh encoding, comprising:

determining (i) a position prediction of a current vertex of a base mesh associated with a mesh in a current mesh frame and (ii) a motion field prediction of the current vertex of the base mesh;

quantizing (i) a position prediction residue for the position prediction of the current vertex to generate a quantized position prediction residue based on a first quantization step value, and (ii) a motion field prediction residue for the motion field prediction of the current vertex to generate a quantized motion field prediction residue based on a second quantization step value, the second quantization step value being determined based on a function of the first quantization step value; and

entropy coding (i) the quantized position prediction residue for the position prediction of the current vertex, and (ii) the quantized motion field prediction residue for the motion field prediction of the current vertex.

14 . The method of claim 13 , wherein the second quantization step value is equal to a multiple of the first quantization step value.

15 . The method of claim 13 , wherein the second quantization step value is equal to one of a linear function and a monotonically non-decreasing function of the first quantization step value.

16 . The method of claim 13 , further comprising:

determining a two-dimensional (2D) texture coordinate prediction for a 2D texture coordinate of the current vertex of the base mesh;

quantizing a 2D texture coordinate prediction residue for a 2D texture coordinate prediction of the current vertex to generate a quantized 2D texture coordinate prediction residue of the current vertex; and

entropy coding the quantized 2D texture coordinate prediction residue of the current vertex.

17 . The method of claim 16 , wherein:

when the 2D texture coordinate of the current vertex is directly coded, the quantizing the 2D texture coordinate prediction residue includes:

quantizing the 2D texture coordinate for the current vertex based on a rounding quantization, the rounding quantization being configured to convert the 2D texture coordinate into an integer; and

coding the quantized 2D texture coordinate based on a fixed-length coding.

18 . The method of claim 16 , wherein:

when the 2D texture coordinate of the current vertex is predicted by a stretch prediction algorithm, the determining the 2D texture coordinate prediction residue includes:

quantizing the 2D texture coordinate prediction residue for the current vertex by rounding a multiple of the 2D texture coordinate prediction residue and an inverse of a dyadic rational of

m being a positive integer between 0 and 7, n being a non-negative integer between 0 and 255; and

coding the quantized 2D texture coordinate prediction residue based on a variable-length coding.

19 . The method of claim 18 , wherein the quantizing the 2D texture coordinate prediction residue further comprises:

adding an offset to the 2D texture coordinate prediction residue to generate an updated 2D texture coordinate prediction residue; and

quantizing the updated 2D texture coordinate prediction residue for the current vertex by rounding a multiple of the updated 2D texture coordinate prediction residue and the inverse of the dyadic rational of

20 . A non-transitory computer-readable storage medium storing instructions which, when executed by a processor, cause the processor to perform a method of encoding a bitstream comprising:

determining (i) a position prediction of a current vertex of a base mesh associated with a mesh in a current mesh frame and (ii) a motion field prediction of the current vertex of the base mesh;

transmitting the encoded bitstream.