【语义分割系列：八】Segmentation 数据集介绍&下载&论文

深度学习语义分割：全面解析数据集

最新推荐文章于 2025-10-14 10:34:45 发布

原创最新推荐文章于 2025-10-14 10:34:45 发布 · 3.2k 阅读

8 ·

CC 4.0 BY-SA版权

文章标签：

#语义分割 #数据集 #最全

图像分割专栏收录该内容

12 篇文章

订阅专栏

本文介绍了多个用于语义分割任务的数据集，包括ADE20K、Cityscapes、PASCAL VOC 2012等。ADE20K包含150类，涵盖室内室外场景；Cityscapes提供50城市的高质像素级图像，有30类；PASCAL VOC 2012则涉及20个对象类别，可用于分类、检测和分割。

部署运行你感兴趣的模型镜像

ADE20K

Link
leaderboard
Pytorch implementation
paper

Train （20000） validation （2000）
test (需要上传结果到服务器才能得到结果)
150类
场景包括室内室外各种各样的场景.

cityscapes

Link
paper
README and scripts

5000 high quality pixel-level image
50 city
30 class

Group	Classes
flat	road · sidewalk · parking+ · rail track+
human	person* · rider*
construction	building · wall · fence · guard rail+ · bridge+ · tunnel+
object	pole · pole group+ · traffic sign · traffic light
nature	vegetation · terrain
sky	sky
void	ground+ · dynamic+ · static+

dataset directory

  |--cityscape
      |--leftImg8bit_trainvaltest
                 |--leftImg8bit (5,000)
                       |--train (2975)
                       |--val   (500)
                       |--test  (1525)
      |--gtFine_trainvaltest    (5000 * 4)
                 |--gtFine
                       |--train (2975 * 4)
                       |--val   (500 * 4)
                       |--test  (1525 * 4)
      |--leftImg8bit_trainextra
                 |--leftImg8bit (20,000)
                       |--train_extra (20000)
      |--gtCoarse
                 |--gtCoarse    (23475 * 4)
                       |--train (2975 * 4)
                       |--train_extra (20000 * 4)
                       |--val   (500 * 4)

pascal VOC

Link
paper

20个对象类别和1个背景类

1 PASCAL VOC2012

Link
Download
leaderboard
paper

classification, detection and segmentation.
20 类
Group| Classes|
-|- |
Person| person|
Animal| bird, cat, cow, dog, horse, sheep|
Vehicle |aeroplane, bicycle, boat, bus, car, motorbike, train|
Indoor |bottle, chair, dining table, potted plant, sofa, tv/monitor|

dataset directory

|--VOCdevkit
      |--VOC2012(trainval)
            |--Annotations (17,125)
                    |--*.xml
            |--ImageSets
                    |--Action (33)
                            |--*.txt
                    |--Layout
                            |--train.txt
                            |--trainval.txt
                            |--val.txt
                    |--Main   (63)
                            |--*.txt
                    |--Segmentation
                            |--train.txt (1464 lines)
                            |--trainval.txt (2913 lines)
                            |--val.txt (1449 lines)
            |--JPEGImages  (17,125)
                    |--*.jpg
            |--SegmentationClass  (2913)
                    |--*.png
            |--SegmentationObject (2913)
                    |--*.png
      |--VOC2012(test)
            |--Annotations (5,138)
                    |--*.xml
            |--ImageSets
                    |--Action (11)
                            |--*.txt
                    |--Layout
                            |--test.txt
                    |--Main   (21)
                            |--*.txt
                    |--Segmentation
                            |--test.txt (1456 lines)
            |--JPEGImages  (16,135)
                    |--*.jpg

2 PASCAL Context

Link
paper
PASCAL VOC 2010

基于PASCAL VOC 2010做的标记
Training and validation （ 10,103） testing （9,637）

3 PASCAL-Part Dataset

Link

MS-COCO

Link
paper

The COCO train, validation, and test sets, containing more than 200,000 images and 80 object categories, are available on the download page.
All object instances are annotated with a detailed segmentation mask.
Annotations on the training and validation sets (with over 500,000 object instances segmented) are publicly available.