本人在使用 Live2D Cubism SDK for Native 加载从github上下载的模型文件时，产生了类似数组越界的问题，为了解决这个方案查找了多个文档，最终在官方的Spec仓库中找到了关于motion3.json的说明，该说明已放在 "官方说明" 文件夹下，可以自行查看.
以下是关于上述数组越界问题的解释:
直接运行官方的SDK for Native加载从网上下载的模型，可能会产生如下错误
导致这个断点，可能原因有：
- 释放了一个未被分配的指针
- 数组访问越界
- 其他对未被分配内存的操作
经过多层debug，最终找到可能导致断点的地方：
把Count参数增大1000，再次运行项目的时候，发现模型可以正常加载
这么粗略的解决肯定是不行的，于是开始关注这几个参数，发现这几个参数对应motion3.json文件中的curveCount，TotalSegmentCount和TotalPointCount等参数。
文件中的这些参数可能有错误，但是这么算这几个参数呢？
经过查找，在网上找到了两条有关这个的说明，结合官方文档和自己实际计算最终得出的计算的方法
如下是一个motion3.json文件
{
"Version": 3,
"Meta": {
"Duration": 6.833,
"Fps": 30.0,
"Loop": true,
"AreBeziersRestricted": true,
"CurveCount": 146,
"TotalSegmentCount": 1115,
"TotalPointCount": 1447,
"UserDataCount": 2,
"TotalUserDataSize": 0
},
"Curves": [
{
"Target": "Parameter",
"Id": "Param_Angle_Rotation_3_ArtMesh53",
"Segments": [
0,
8.956,
2,
0.333,
8.956,
1,
0.389,
8.956,
0.444,
7.438,
0.5,
5.129,
1,
0.756,
-5.489,
1.011,
-10.482,
1.267,
-10.482,
0,
1.767,
11.301,
0,
2.2,
-17.088,
0,
2.6,
9.684,
0,
3.033,
-10.562,
0,
3.267,
-3.355,
0,
3.6,
-10.369,
0,
4.1,
10.391,
0,
4.467,
-13.114,
0,
5.1,
5.97,
0,
5.633,
-11.428,
0,
6.333,
8.956,
1,
6.422,
8.956,
6.511,
7.597,
6.6,
5.129,
1,
6.678,
2.97,
6.755,
1.038,
6.833,
-0.668
]
},
{
"Target": "Parameter",
"Id": "Param_Angle_Rotation_4_ArtMesh53",
"Segments": [
0,
7.009,
2,
0.333,
7.009,
0,
0.6,
7.017,
0,
1.3,
-12.174,
0,
1.9,
7.293,
0,
2.367,
-13.252,
0,
2.767,
8.575,
0,
3.167,
-6.631,
0,
3.533,
-1.717,
0,
3.767,
-10.758,
0,
4.233,
6.945,
0,
4.667,
-11.214,
0,
5.1,
5.668,
0,
5.8,
-7.739,
1,
5.978,
-7.739,
6.155,
6.995,
6.333,
7.009,
1,
6.478,
7.021,
6.622,
7.017,
6.767,
7.017,
1,
6.789,
7.017,
6.811,
6.937,
6.833,
6.788
]
}
]
}一个 Curve 就是 Curves 数组中的一个元素
Curve 中的 Segments 数组解析如下:
- 一个 point 包含两个浮点数
- 一个 segment 可以等价为一个 identifier [1] (表示 segment 类型的整数) 与1个或3个 point 构成的集合
- 一个 Segments 数组最开始的两个数代表一个 point
- 于是 Segments 数组 [num1, num2, num3, num4, num5, num6, num7, num8, ...] 等价于如下定义
[
Point(num1, num2), # 开头两个数组成一个单独的point，它们之后是一系列segment
Segment(
Identifier1(num3), # 第三个数是 Identifier 表示 Segment 类型
Point(num4, num5) # Identifier 之后是1个或3个 point
),
Segment(
Identifier(num6),
Point(num7, num8)
)
...
]解析完毕之后就是写代码计算了
Curves are made up of points and segment identifiers packed into a single flat array.
A point is a sequence of 2 numbers, where the first number is its timing in seconds, hereinafter simply t, and the second number its value at t.
A segment identifier is single number with one of the following values.
| Value | Description |
|---|---|
| 0 | Identifier for linear segment. |
| 1 | Identifier for cubic bézier segment. |
| 2 | Identifier for stepped segment |
| 3 | Identifier for inverse-stepped segment. |
Each curve starts with the first point followed by the segment identifier. Therefore, the first segment identifier is the third number in the flat segments array. Segments can be reconstructed by taking the point before the segment identifier and combining it with the points until the next identifier (or until the end of the array). A segment identifier is followed by 1 point in case of linear, stepped, and inverse stepped segments, that represents the end of the segment, or 3 point in case of bézier segments, that represent P1, P2, P3.
The t components of bézier control points are proportionally fixed as follows.
Curves can't be empty.
- Python 3.10及以上
- 无
项目中的motion_spec.py是核心工具，下面介绍它的用法
自己创建一个main.py的文件 然后把motion_spec.py放在和main.py同级的文件夹下 项目结构
project
|
|--motion_spec.py
|
|--main.py
|
|---Hiyori
|
|--motions
|
|--Hiyori_01.motion3.json
|
...
main.py
from motion_spec import copy_modify_from_motion
motion_path = "Hiyori/motions/Hiyori_01.motion3.json" #模型文件的路径
save_path = "out/motions" #文件的输出文件夹
copy_modify_from_motion(motion_path, save_path)# motion3.json文件路径列表
model_dir = "xuefeng_3" # 模型文件夹路径
out_path = "./out/motions" # 输出文件夹路径
motionPathList = load_all_motion_path_from_model_dir(model_dir)
for path in motionPathList:
# path 是motion3.json文件所在路径
copy_modify_from_motion(path, save_root=out_path)模型文件夹应该有如下内容
