IDD-3D Indian Driving Dataset for 3D Unstructured Road Scenes Shubham Dokania1 A. H. Abdul Hafez2 Anbumani Subramanian1 Manmohan Chandraker3 C.V . Jawahar1

2025-05-08 11 0 4.61MB 10 页 10玖币

侵权投诉

IDD-3D: Indian Driving Dataset for 3D Unstructured Road Scenes

Shubham Dokania1, A. H. Abdul Hafez2, Anbumani Subramanian1,

Manmohan Chandraker3, C.V. Jawahar1

1IIIT Hyderabad, 2Hasan Kalyoncu University, 3UC San Diego

shubham.dokania@research.iiit.ac.in, abdul.hafez@hku.edu.tr,

anbumani@iiit.ac.in, mkchandraker@eng.ucsd.edu, jawahar@iiit.ac.in

Abstract

Autonomous driving and assistance systems rely on an-

notated data from trafﬁc and road scenarios to model and

learn the various object relations in complex real-world

scenarios. Preparation and training of deploy-able deep

learning architectures require the models to be suited to

different trafﬁc scenarios and adapt to different situations.

Currently, existing datasets, while large-scale, lack such di-

versities and are geographically biased towards mainly de-

veloped cities. An unstructured and complex driving layout

found in several developing countries such as India poses a

challenge to these models due to the sheer degree of varia-

tions in the object types, densities, and locations. To facili-

tate better research toward accommodating such scenarios,

we build a new dataset, IDD-3D, which consists of multi-

modal data from multiple cameras and LiDAR sensors with

12k annotated driving LiDAR frames across various traf-

ﬁc scenarios. We discuss the need for this dataset through

statistical comparisons with existing datasets and highlight

benchmarks on standard 3D object detection and tracking

tasks in complex layouts. Code and data available 1.

1. Introduction

Intelligent vehicles and autonomous driving systems

have come a long way and keep becoming more sophisti-

cated over time, owing to the rapid progress in the deep

learning and computer vision. However, the core compo-

nent for all these increments is the availability of high-

quality annotated data. Recently, many works have fo-

cused on data selection and quality improvement [34, 8,

47], building high-quality and large-scale datasets, and ap-

proaches built using these resources, which improve the

state of autonomous driving [48, 16].

Existing datasets are usually collected in well-structured

environments with proper trafﬁc regulations and relatively-

1https://github.com/shubham1810/idd3d_kit.git

Figure 1. Some examples from the dataset showing different trafﬁc

scenarios, LiDAR data with annotations, and a sample of LiDAR

point clouds projected on camera data.

evenly distributed trafﬁc. In such situations, crowd behavior

demonstrates low diversity and average densities. In south-

east Asian countries, such as India, the trafﬁc densities and

inter-object behaviors are much more complex. Such com-

plexities have been studied in the past [39, 5, 4], but ex-

tensive data coverage and multi-modal systems are still un-

available for such scenes. It hence may not be entirely ap-

plied to cases where the distribution of object categories and

types varies greatly.

In this paper, we propose a dataset on complex unstruc-

tured driving scenarios with multi-modal data, highlighting

the capabilities of 3D sensors such as LiDAR for better

scene perception in unstructured and sporadically chaotic

trafﬁc conditions. In the proposed dataset, we highlight a

signiﬁcantly different distribution of object types and cat-

egories compared to existing datasets collected in Euro-

pean or similar settings [24, 13, 38], due to the different

nature of trafﬁc scenes in Indian roads. Furthermore, the

categories and annotations available in the proposed dataset

vary greatly from existing datasets. Speciﬁcally, they cover

objects in scenes that usually appear in still-developing

arXiv:2210.12878v1 [cs.CV] 23 Oct 2022

Figure 2. Samples from the dataset highlighting different (a) RGB images and (b) LiDAR Bird-Eye-View (BEV) along with bounding box

annotations. The samples visualized above are taken from different sequences of the dataset.

cities, for example, Auto-rickshaws, hand carts, concrete

mixer machines on roads, and animals on roads.

We provide data collected in Indian road scenes, from

high-quality LiDAR sensors and six cameras that cover

the surrounding area of the ego-vehicle to enable sensor-

fusion-based applications. We provide annotations for

15.5k frames in the dataset, which spans 10 primary cat-

egories (and 7 additional miscellaneous categories), which

we use for model training and evaluation. Along with the

annotations, we also provide extra unlabelled raw data from

the sensors to facilitate further research, especially into

self- and unsupervised learning over such trafﬁc scenes. A

unique feature of the proposed dataset, which stems from

the unstructured environment, is the availability of highly

complex trajectories. We show samples from the dataset

which emphasize such cases and display experiments on ob-

ject detection and tracking, which is possible due to avail-

ability of instance speciﬁc labels for each object bounding

box per sequence.

Our main contributions can be summarised as follows:

(i) We propose the IDD-3D dataset for driving in unstruc-

tured trafﬁc scenarios for Indian roads with 3D informa-

tion, (ii) high-quality annotations for 3D object bounding

boxes with 9DoF data, and instance IDs to enable tracking,

(iii) Analysis over highly unstructured and diverse environ-

ments to accentuate the usefulness of proposed dataset, and

(iv) provide 3D object detection and tracking benchmarks

across popular methods in literature.

2. Related Work

Data plays a huge role in machine learning systems, and

in this context, for autonomous vehicles and scene percep-

tion. There have been several efforts over the years in this

area to improve the state of datasets available and towards

increasing the volumes of high-quality and well annotated

datasets.

2D Driving: One of the early datasets towards visual per-

ception and understanding driving has been the CamVid

[2] and Cityscapes [9, 10] dataset, providing annotations

for semantic segmentation and enabling research in deeper

scene understanding at pixel-level. KITTI [14, 15] dataset

provided 2D object annotations for detection and tracking

along with segmentation data. However, fusion of multi-

ple modalities such as 3D LiDAR data enhances the perfor-

mance for scene understanding benchmarks as these pro-

vide a higher level of detail of a scene when combined with

available 2D data. This multi-modal sensor-fusion based

direction has been the motivation for the proposed dataset

to alleviate the discrepancies in existing datasets for scene

文档加载中……请稍候！
如果长时间未打开，您也可以点击刷新试试。

下载文档到电脑，查找使用更方便

10 玖币 0人已下载

立即下载

摘要：

IDD-3D:IndianDrivingDatasetfor3DUnstructuredRoadScenesShubhamDokania1,A.H.AbdulHafez2,AnbumaniSubramanian1,ManmohanChandraker3,C.V.Jawahar11IIITHyderabad,2HasanKalyoncuUniversity,3UCSanDiegoshubham.dokania@research.iiit.ac.in,abdul.hafez@hku.edu.tr,anbumani@iiit.ac.in,mkchandraker@eng.ucsd.edu,jawah...

展开>> 收起<<

IDD-3D Indian Driving Dataset for 3D Unstructured Road Scenes Shubham Dokania1 A. H. Abdul Hafez2 Anbumani Subramanian1 Manmohan Chandraker3 C.V . Jawahar1.pdf

共10页,预览2页

还剩页未读，继续阅读

声明：本站为文档C2C交易模式，即用户上传的文档直接被用户下载，本站只是中间服务平台，本站所有文档下载所得的收益归上传人(含作者)所有。玖贝云文库仅提供信息存储空间，仅对用户上传内容的表现方式做保护处理，对上载内容本身不做任何修改或编辑。若文档所含内容侵犯了您的版权或隐私，请立即通知玖贝云文库，我们立即给予删除！

IDD-3D Indian Driving Dataset for 3D Unstructured Road Scenes Shubham Dokania1 A. H. Abdul Hafez2 Anbumani Subramanian1 Manmohan Chandraker3 C.V . Jawahar1

相关推荐

开通VIP享超值会员特权

作者详情

相关内容

热门标签

举报选择: