---
title: NeuralConvection-2D-TrainTest-v1
canonical_url: "https://www.modelscope.cn/datasets/OneScience/NeuralConvection-2D-TrainTest-v1"
md_url: "https://www.modelscope.cn/datasets/OneScience/NeuralConvection-2D-TrainTest-v1.md"
repository: OneScience/NeuralConvection-2D-TrainTest-v1
last_updated: 2026-10-18
license: cc-by-4.0
storage_size: "575 MB"
downloads: 4
stars: 0
---

# NeuralConvection-2D-TrainTest-v1

> NeuralConvection-2D-TrainTest-v1 - OneScience 在 ModelScope 开源的数据集。NeuralConvection 2D Train/Test v1

OneScience/NeuralConvection-2D-TrainTest-v1 是 ModelScope 魔搭社区上的数据集，存储大小 575 MB，采用 cc-by-4.0 许可。

- **Repository**: OneScience/NeuralConvection-2D-TrainTest-v1
- **License**: cc-by-4.0
- **Storage size**: 575 MB
- **Downloads**: 4
- **Stars**: 0
- **Last updated**: 2026-10-18

Source: https://www.modelscope.cn/datasets/OneScience/NeuralConvection-2D-TrainTest-v1

---

# NeuralConvection 2D Train/Test v1

## 数据集介绍

本数据集整理自[作者公开的 NeuralConvection 2D 数据集](https://huggingface.co/datasets/NurshatMenglik/NeuralConvection_2D_TrainTest/tree/v1.0-legacy)的 `v1.0-legacy` 固定版本，用于二维自然对流模型实验。文件与该版本逐个 SHA256 一致，未与后来发布的 v2 数据混用。

## 数据集支持的任务

本数据集可用于 [OneScience/NeuralConvection-2D-CompactFNO](https://www.modelscope.cn/models/OneScience/NeuralConvection-2D-CompactFNO) 的训练与评估：从 Boussinesq 模拟场预测对应的可压缩流场。训练集有 4000 组；同分布、分布外、传统恒温边界三个测试集各有 200 组。

## 数据格式

| 目录 | 样本数 | 文件 | shape | dtype |
| --- | ---: | --- | --- | --- |
| `train/` | 4000 | `X.pt`, `Y.pt` | `[4000,4,64,64]` | `float32` |
| `test_in_distribution/` | 200 | `X.pt`, `Y.pt` | `[200,4,64,64]` | `float32` |
| `test_ood/` | 200 | `X.pt`, `Y.pt` | `[200,4,64,64]` | `float32` |
| `test_legacy/` | 200 | `X.pt`, `Y.pt` | `[200,4,64,64]` | `float32` |

`X.pt` 为 Boussinesq 输入，`Y.pt` 为可压缩目标；通道顺序为 `[u_x,u_y,T,p]`。

## 数据集使用方式

在模型包根目录执行：

```bash
pip install 'modelscope[datasets]'
modelscope download --dataset OneScience/NeuralConvection-2D-TrainTest-v1 --local_dir ./data
```

## OneScience 官方信息

| 平台 | OneScience 主仓库 | Skills 仓库 |
| --- | --- | --- |
| Gitee | https://gitee.com/onescience-ai/onescience | https://gitee.com/onescience-ai/oneskills |
| GitHub | https://github.com/onescience-ai/OneScience | https://github.com/onescience-ai/oneskills |

## 引用与许可证

- 论文：[A Neural Surrogate Approach for Simulating Natural Convection Problems](https://arxiv.org/abs/2606.25259)。
- 原始数据：[NeuralConvection 2D Train/Test，`v1.0-legacy`](https://huggingface.co/datasets/NurshatMenglik/NeuralConvection_2D_TrainTest/tree/v1.0-legacy)，作者标注为 CC BY 4.0；使用和再分发时须保留来源与署名。