Sta8is/cityscapes_segmenter_ids

---
license: mit
language:
- en
tags:
- Semantic_Segmentation
size_categories:
- 10K<n<100K
---

# Dataset Card for Cityscapes Semantic Segmentation ids from Segmenter

This dataset contains precomputed semantic segmentation maps (label IDs) for the Cityscapes dataset, generated using a [Segmenter](https://github.com/rstrudel/segmenter) model.
---
## Dataset Details
### Dataset Description
This dataset provides precomputed **per-pixel semantic labels** (IDs) derived from Cityscapes images using a pretrained segmentation model. Each image is stored as a single-channel PNG, where pixel values correspond to Cityscapes training IDs (0–18), with `255` as the ignore label.

### Dataset Sources
- **Repository:** https://huggingface.co/datasets/Sta8is/cityscapes_segmenter_ids
- **Project page:** https://futurist-cvpr2025.github.io/
---
## Uses
### Direct Use
- Semantic future prediction
- Multimodal learning pipelines
- Efficient experimentation without recomputing segmentation
### Out-of-Scope Use
- Not suitable as ground-truth labels
- Not intended for benchmarking segmentation accuracy
---
## Dataset Structure
The dataset follows the standard Cityscapes split:

leftImg8bit_sequence_segmaps_ids/
train/
val/
test/

- Files are single-channel `.png` images
- Pixel values correspond to class IDs (0–18)
- Ignore label: 255

---
## Dataset Creation
### Curation Rationale
Created to reduce computational overhead and improve reproducibility in research workflows that use segmentation as input.
### Source Data
Derived from the Cityscapes sequence dataset using a pretrained segmentation model.

---
## Bias, Risks, and Limitations
- Labels are **model-generated** and may contain errors
- Inherits biases from Cityscapes and the segmentation model
### Recommendations
Use for input representations, not as evaluation ground truth.

---
## Citation
**BibTeX:**

@InProceedings{Karypidis_2025_CVPR,
author = {Karypidis, Efstathios and Kakogeorgiou, Ioannis and Gidaris, Spyros and Komodakis, Nikos},
title = {Advancing Semantic Future Prediction through Multimodal Visual Sequence Transformers},
booktitle = {CVPR},
year = {2025}
}

---
## Dataset Card Contact
Efstathios Karypidis
[e-mail]

许可证：MIT许可证
语言：英语
标签：语义分割（Semantic_Segmentation）
样本量级：1万 < 样本数 < 10万

# 基于Segmenter模型的Cityscapes语义分割ID数据集卡片

本数据集包含基于Cityscapes数据集的预计算语义分割图（标签ID），由Segmenter模型生成。

## 数据集详情
### 数据集描述
本数据集提供经预计算的**逐像素语义标签（ID）**，通过预训练分割模型从Cityscapes图像中提取得到。所有图像均以单通道PNG格式存储，像素值对应Cityscapes训练类别ID（0~18），其中像素值255为忽略标签。

### 数据集来源
- **代码仓库：** https://huggingface.co/datasets/Sta8is/cityscapes_segmenter_ids
- **项目主页：** https://futurist-cvpr2025.github.io/

## 用途
### 直接用途
- 语义未来预测
- 多模态学习流水线
- 无需重复计算分割结果即可开展高效实验

### 不适用场景
- 不可作为真实标注（Ground Truth）使用
- 不用于分割精度的基准测试

## 数据集结构
本数据集遵循标准Cityscapes数据集划分方式：

leftImg8bit_sequence_segmaps_ids/
train/
val/
test/

- 数据文件均为单通道`.png`图像
- 像素值对应类别ID（0~18）
- 忽略标签为255

## 数据集构建
### 构建初衷
本数据集旨在减少以分割结果作为输入的研究工作流中的计算开销，并提升研究可复现性。

### 源数据
本数据集通过预训练分割模型从Cityscapes序列数据集提取得到。

## 偏差、风险与局限性
- 标签为**模型生成**，可能存在错误
- 继承自Cityscapes数据集与分割模型的固有偏差

### 使用建议
仅可将其作为输入表征使用，不可作为评估用真实标注。

## 引用信息
**BibTeX格式引用：**
bibtex
@InProceedings{Karypidis_2025_CVPR,
author = {Karypidis, Efstathios and Kakogeorgiou, Ioannis and Gidaris, Spyros and Komodakis, Nikos},
title = {基于多模态视觉序列Transformer推进语义未来预测},
booktitle = {CVPR},
year = {2025}
}

## 数据集卡片联系人
埃夫斯塔西奥斯·卡里皮季斯（Efstathios Karypidis）
[e-mail]