Dataset Ninja LogoDataset Ninja:

IDD: Segmentation Dataset

20101411934
Tagself-driving, benchmark
Taskinstance segmentation
Release YearMade in 2018
Licenseunknown

Introduction #

Girish Varma, Anbumani Subramanian, Anoop Namboodiriet al.

The authors of India Driving Dataset (IDD): A Dataset for Exploring Problems of Autonomous Navigation in Unconstrained Environments highlight a notable gap in existing datasets, which primarily focus on structured driving environments with well-defined infrastructure, limited traffic categories, and adherence to traffic rules. To fill this void, the authors present IDD, a novel dataset tailored for road scene understanding in unstructured environments, specifically on Indian roads. The updated version of the dataset (acquired in Oct, 2023) comprises 20k images, meticulously annotated with 41 classes, derived from 182 drive sequence.

There are 2 different versions available at DatasetNinja:

IDD diverges from popular benchmarks like Cityscapes, introducing an expanded label set to accommodate new classes and reflecting label distributions that deviate significantly from existing datasets. The dataset captures the complexity of unstructured road scenes, featuring classes with greater within-class diversity. Additionally, IDD identifies new classes such as drivable areas beyond the road. The authors propose a four-level label hierarchy (level1id, level2id, level3id, level4id) to allow varying levels of complexity, opening avenues for new training methods. Please note, that some labels were not specified by their category

image

Label distribution in the initial IDD-10k dataset. The following information is shown here: (i) pixel counts of individual labels on the y-axis (ii) four-level label hierarchy used by the dataset at the bottom, (iv) the color legend for the predicted and ground truth masks shown in the paper is used for the corresponding bars. There are 4 levels of the hierarchy giving different complexity levels for training models

Empirical studies demonstrate that state-of-the-art methods for semantic segmentation exhibit lower accuracies on IDD compared to Cityscapes (available on DatasetNinja), emphasizing its distinct characteristics. The authors suggest that IDD presents an ideal opportunity for exploring new challenges such as domain adaptation, few-shot learning, and behavior prediction in road scenes.

Autonomous navigation is rapidly advancing, and the availability of large-scale datasets is a crucial contributor to this progress. However, challenges persist, particularly in achieving data scale and diversity necessary for ensuring safety and reliability in diverse and unstructured environments. The authors assert that IDD, with its focus on less structured environments, better aligns with the needs of autonomous navigation in various parts of the world, including Asia, South America, and Africa.

Dataset properties

The dataset is collected in India, offering road scenes that differ significantly from those in Europe or North America. Indian roads exhibit a greater variety of traffic participants, including unique classes such as autorickshaws and animals. Within-class diversity is higher due to variations in vehicle manufacturing years and wear. The dataset’s distribution of classes, even those overlapping with Cityscapes, differs significantly, underscoring the need for a more diverse and complex dataset.

IDD incorporates extensive variations in ambient conditions, such as lighting, shadows, clouded skies, and particulate matter, further increasing its complexity. The dataset is annotated at a scale significantly larger than contemporary datasets like KITTI or Cityscapes, providing high-quality annotations for 41 labels.

The authors conduct a thorough analysis of the dataset, pointing out challenges observed when applying models trained on existing datasets. They emphasize ambiguous road boundaries, diversity in vehicles and pedestrians, extensive use of information boards, and variations in ambient conditions as distinctive features of unstructured environments.

Dataset acquisition, pre-processing and statistics

The dataset, acquired from Bangalore and Hyderabad cities in India, comprises 182 drive sequences with a mix of urban and rural areas, highways, single and double-lane roads. The driving conditions are highly unstructured due to the following reasons:

  1. These cities are experiencing rapid growth, marked by extensive construction near roads.
  2. Additionally, road boundaries lack clear definition.
  3. Pedestrians and jaywalkers are abundant in road images.
  4. A high density of motorbikes and trucks on the roads. Moreover, there is considerable diversity in vehicle models.

Images are selected from forward-facing cameras, densely sampled around crowded and special interest places. The dataset is split into 70% train, 10% validation, and 20% test, with careful consideration for class imbalance.

image

Comparison of the pixel count in our dataset with Cityscapes. The y axis is plotted in log-scale. Note that for most classes of vehicles, the number of pixels are 5-10 times more than Cityscapes. Moreover our dataset has newer labels like autorickshaw, billboard, drivable/nondrivable fallback which also have significant number of labeled pixels.

Comparisons with other datasets, such as Cityscapes, KITTI, and Mapillary Vistas, highlight the unique characteristics of IDD. The authors present detailed label statistics, pixel counts, and instances of traffic participants, demonstrating the dataset’s distinctions.

image

Comparison of traffic participants in our dataset with Cityscapes.

In conclusion, IDD stands out as a comprehensive and challenging dataset for unstructured road scene understanding, offering new opportunities for advancing research in autonomous navigation and related domains.

ExpandExpand
Dataset LinkHomepageDataset LinkResearch PaperDataset LinkGitHub

Summary #

India Driving Dataset (IDD): A Dataset for Exploring Problems of Autonomous Navigation in Unconstrained Environments (Segmentation 20k) is a dataset for instance segmentation, semantic segmentation, and object detection tasks. It is used in the automotive industry.

The dataset consists of 20101 images with 1519090 labeled objects belonging to 41 different classes including obs-str-bar-fallback, sky, vegetation, and other: pole, road, drivable fallback, building, motorcycle, car, rider, billboard, person, wall, vehicle fallback, non-drivable fallback, autorickshaw, curb, truck, traffic sign, bus, sidewalk, fence, fallback background, guard rail, bridge, polegroup, bicycle, animal, traffic light, out of roi, caravan, parking, tunnel, rail track, trailer, rectification border, train, ego vehicle, license plate, ground, and unlabeled.

Images in the IDD: Segmentation dataset have pixel-level instance segmentation annotations. Due to the nature of the instance segmentation task, it can be automatically transformed into a semantic segmentation (only one mask for every class) or object detection (bounding boxes for every object) tasks. There are 4050 (20% of the total) unlabeled images (i.e. without annotations). There are 3 splits in the dataset: train (14027 images), test (4038 images), and val (2036 images). Additionally, every image has information about its sequence, while every object instance has its category and 4-level label hierarchy: level1id, level2id, level3id, level4id. Explore them in supervisely labeling tool. The dataset was released in 2018 by the IIIT Hyderabad, India, Intel Bangalore, and University of California, San Diego.

Here is a visualized example for randomly selected sample classes:

Explore #

IDD: Segmentation dataset has 20101 images. Click on one of the examples below or open "Explore" tool anytime you need to view dataset images with annotations. This tool has extended visualization capabilities like zoom, translation, objects table, custom filters and more. Hover the mouse over the images to hide or show annotations.

OpenSample annotation mask from IDD: SegmentationSample image from IDD: Segmentation
OpenSample annotation mask from IDD: SegmentationSample image from IDD: Segmentation
OpenSample annotation mask from IDD: SegmentationSample image from IDD: Segmentation
OpenSample annotation mask from IDD: SegmentationSample image from IDD: Segmentation
OpenSample annotation mask from IDD: SegmentationSample image from IDD: Segmentation
OpenSample annotation mask from IDD: SegmentationSample image from IDD: Segmentation
OpenSample annotation mask from IDD: SegmentationSample image from IDD: Segmentation
OpenSample annotation mask from IDD: SegmentationSample image from IDD: Segmentation
OpenSample annotation mask from IDD: SegmentationSample image from IDD: Segmentation
OpenSample annotation mask from IDD: SegmentationSample image from IDD: Segmentation
OpenSample annotation mask from IDD: SegmentationSample image from IDD: Segmentation
OpenSample annotation mask from IDD: SegmentationSample image from IDD: Segmentation
👀
Have a look at 20101 images
Because of dataset's license preview is limited to 12 images
View images along with annotations and tags, search and filter by various parameters

Class balance #

There are 41 annotation classes in the dataset. Find the general statistics and balances for every class in the table below. Click any row to preview images that have labels of the selected class. Sort by column to find the most rare or prevalent classes.

Search
Rows 1-10 of 41
Class
ã…¤
Images
ã…¤
Objects
ã…¤
Count on image
average
Area on image
average
unlabeledâž”
polygon
16976
16976
1
23.88%
obs-str-bar-fallbackâž”
polygon
15966
322213
20.18
5.25%
skyâž”
polygon
15929
16489
1.04
35.58%
vegetationâž”
polygon
15914
285290
17.93
19.84%
poleâž”
polygon
15751
185634
11.79
1.32%
roadâž”
polygon
15708
16475
1.05
39.07%
drivable fallbackâž”
polygon
14072
30190
2.15
10.88%
buildingâž”
polygon
14006
64637
4.61
16%
motorcycleâž”
polygon
13594
83046
6.11
1.72%
carâž”
polygon
13273
69929
5.27
4.1%

Images #

Explore every single image in the dataset with respect to the number of annotations of each class it has. Click a row to preview selected image. Sort by any column to find anomalies and edge cases. Use horizontal scroll if the table has many columns for a large number of classes in the dataset.

Class sizes #

The table below gives various size properties of objects for every class. Click a row to see the image with annotations of the selected class. Sort columns to find classes with the smallest or largest objects or understand the size differences between classes.

Search
Rows 1-10 of 41
Class
Object count
Avg area
Max area
Min area
Min height
Min height
Max height
Max height
Avg height
Avg height
Min width
Min width
Max width
Max width
obs-str-bar-fallback
polygon
322213
0.27%
67.82%
0%
1px
0.09%
1080px
100%
56px
5.42%
2px
0.1%
1920px
100%
vegetation
polygon
285290
1.28%
56.99%
0%
1px
0.09%
1080px
100%
136px
13.32%
2px
0.1%
1920px
100%
pole
polygon
185634
0.1%
49.66%
0%
2px
0.19%
1080px
100%
149px
14.49%
1px
0.05%
1920px
100%
billboard
polygon
100708
0.27%
17.71%
0%
4px
0.37%
992px
91.85%
62px
6%
3px
0.16%
1726px
98.52%
motorcycle
polygon
83046
0.29%
99.86%
0%
3px
0.28%
1080px
100%
76px
7.39%
3px
0.16%
1920px
100%
rider
polygon
73306
0.19%
42.39%
0%
3px
0.28%
739px
85.17%
81px
7.94%
3px
0.16%
1920px
100%
car
polygon
69929
0.8%
41.59%
0%
3px
0.28%
912px
90.5%
79px
8.06%
3px
0.16%
1792px
93.33%
building
polygon
64637
3.6%
75.46%
0%
4px
0.37%
1080px
100%
199px
19.5%
3px
0.21%
1920px
100%
person
polygon
60617
0.18%
10.73%
0%
2px
0.19%
783px
93.67%
87px
8.42%
2px
0.1%
556px
28.96%
vehicle fallback
polygon
35422
0.17%
43.09%
0%
2px
0.19%
1078px
99.81%
35px
3.46%
3px
0.16%
1920px
100%

Spatial Heatmap #

The heatmaps below give the spatial distributions of all objects for every class. These visualizations provide insights into the most probable and rare object locations on the image. It helps analyze objects' placements in a dataset.

Spatial Heatmap

Objects #

Table contains all 100644 objects. Click a row to preview an image with annotations, and use search or pagination to navigate. Sort columns to find outliers in the dataset.

Search
Rows 1-10 of 100644
Object ID
ã…¤
Class
ã…¤
Image name
click row to open
Image size
height x width
Height
ã…¤
Height
ã…¤
Width
ã…¤
Width
ã…¤
Area
ã…¤
1âž”
sky
polygon
155_156855_leftImg8bit.png
1080 x 1920
328px
30.37%
821px
42.76%
9.94%
2âž”
road
polygon
155_156855_leftImg8bit.png
1080 x 1920
620px
57.41%
1920px
100%
45.97%
3âž”
drivable fallback
polygon
155_156855_leftImg8bit.png
1080 x 1920
506px
46.85%
847px
44.11%
4.25%
4âž”
sidewalk
polygon
155_156855_leftImg8bit.png
1080 x 1920
472px
43.7%
677px
35.26%
7.14%
5âž”
non-drivable fallback
polygon
155_156855_leftImg8bit.png
1080 x 1920
42px
3.89%
154px
8.02%
0.18%
6âž”
building
polygon
155_156855_leftImg8bit.png
1080 x 1920
328px
30.37%
213px
11.09%
2.47%
7âž”
building
polygon
155_156855_leftImg8bit.png
1080 x 1920
28px
2.59%
65px
3.39%
0.07%
8âž”
wall
polygon
155_156855_leftImg8bit.png
1080 x 1920
11px
1.02%
22px
1.15%
0.01%
9âž”
vegetation
polygon
155_156855_leftImg8bit.png
1080 x 1920
65px
6.02%
75px
3.91%
0.14%
10âž”
vegetation
polygon
155_156855_leftImg8bit.png
1080 x 1920
109px
10.09%
82px
4.27%
0.25%

License #

License is unknown for the India Driving Dataset (IDD): A Dataset for Exploring Problems of Autonomous Navigation in Unconstrained Environments (Segmentation 20k) dataset.

Source

Citation #

If you make use of the IDD: Segmentation data, please cite the following reference:

@dataset{IDD: Segmentation,
  author={Girish Varma and Anbumani Subramanian and Anoop Namboodiri and Manmohan Chandraker and C. V. Jawahar},
  title={India Driving Dataset (IDD): A Dataset for Exploring Problems of Autonomous Navigation in Unconstrained Environments (Segmentation 20k)},
  year={2018},
  url={http://idd.insaan.iiit.ac.in/}
}

Source

If you are happy with Dataset Ninja and use provided visualizations and tools in your work, please cite us:

@misc{ visualization-tools-for-idd-segmentation-dataset,
  title = { Visualization Tools for IDD: Segmentation Dataset },
  type = { Computer Vision Tools },
  author = { Dataset Ninja },
  howpublished = { \url{ https://datasetninja.com/idd-segmentation } },
  url = { https://datasetninja.com/idd-segmentation },
  journal = { Dataset Ninja },
  publisher = { Dataset Ninja },
  year = { 2024 },
  month = { jul },
  note = { visited on 2024-07-27 },
}

Download #

Please visit dataset homepage to download the data.

. . .

Disclaimer #

Our gal from the legal dep told us we need to post this:

Dataset Ninja provides visualizations and statistics for some datasets that can be found online and can be downloaded by general audience. Dataset Ninja is not a dataset hosting platform and can only be used for informational purposes. The platform does not claim any rights for the original content, including images, videos, annotations and descriptions. Joint publishing is prohibited.

You take full responsibility when you use datasets presented at Dataset Ninja, as well as other information, including visualizations and statistics we provide. You are in charge of compliance with any dataset license and all other permissions. You are required to navigate datasets homepage and make sure that you can use it. In case of any questions, get in touch with us at hello@datasetninja.com.