ShapeNetCar / metadata /schema.yaml
Zhongning's picture
Upload folder using huggingface_hub (part 17)
a207665 verified
Raw
History Blame Contribute Delete
1.78 kB
dataset: ShapeNetCar
root: data/mlcfd_data
format:
- npy
- vtk
- txt
- xlsx
- png
statistics:
stats/mean_in.npy:
shape: [7]
dtype: float32
description_zh: 输入特征归一化均值
stats/std_in.npy:
shape: [7]
dtype: float32
description_zh: 输入特征归一化标准差
stats/mean_out.npy:
shape: [4]
dtype: float32
description_zh: 输出物理量归一化均值
stats/std_out.npy:
shape: [4]
dtype: float32
description_zh: 输出物理量归一化标准差
training_data:
directory_pattern: training_data/param*
required_files:
Cd.npy: 阻力系数数组
I1.npy: 输入参数数组
I2.npy: 输入参数数组
Press.npy: 表面压力数组
Velo.npy: 速度场数组
topo_hexquad.txt: 网格拓扑文本
preprocessed_data:
directory_pattern: preprocessed_data/param*/<sample_id>
required_files:
x.npy:
shape: [num_nodes, 7]
dtype: float64
description_zh: 节点输入特征,包含位置、SDF 和法向量等特征
y.npy:
shape: [num_nodes, 4]
dtype: float64
description_zh: 目标物理场,前三维为速度,最后一维为压力
pos.npy:
shape: [num_nodes, 3]
dtype: float32
description_zh: 节点三维坐标
surf.npy:
shape: [num_nodes]
dtype: float64
description_zh: 表面节点掩码
edge_index.npy:
shape: [2, num_edges]
dtype: int64
description_zh: 图边索引
notes_zh:
- training_data/param0 param8 对应不同参数折。
- preprocessed_data 中的样本目录由车辆几何样本 ID 命名。
- Transolver-Car-Design 标准模型包默认读取 data/mlcfd_data/training_data、data/mlcfd_data/preprocessed_data data/mlcfd_data/stats。