SVLD-3D Dataset

Introduction

Surveillance Vehicle Localization Dataset (SVLD-3D) - 3D vehicle detection and localization for monocular roadside cameras.

• 5 scenes with 3 typical vehicle types.
• 14592 train/val images, 2237 test images.
• Examples:

Directory Structure

SVLD-3D                             # root directory
    ├── DATA2021                    # trainval set
        ├── Annotations             # annotation
            ├── {image_id}.xml
        ├── Calib                   # calibration parameters (world to image)
            ├── {scene_id}.xml
        ├── ImageSets               # dataset split
            ├── Main
                ├── trainval.txt
                ├── train.txt
                ├── val.txt
                ├── test.txt
        ├── JPEGImages              # image (jpg/png)
            ├── {image_id}.jpg/png
    ├── TESTDATA2021                # test set
        ├── Annotations             # annotation
            ├── {image_id}.xml
        ├── Calib                   # calibration parameters (world to image)
            ├── {scene_id}.xml
        ├── JPEGImages              # image (jpg/png)
            ├── {image_id}.jpg/png
    ├── split_train_val_test.py     # split train/val/test set
    ├── README.md                   # instruction

Annotation Format

<annotation>
	<filename>absolute path: {image_id}.jpg/png</filename>
	<calibfile>absolute path: {scene_id}.xml</calibfile>
	<size>
		<width>image_width</width>
		<height>image_height</height>
		<depth>image_depth</depth>
	</size>
	<object>
		<type>vehicle type</type>
		<bbox2d>2D bbox: left, top, right, bottom</bbox2d>
		<vertex2d>3D bbox(image): pt1~pt8</vertex2d>
		<veh_size>3D bbox dimension(m)</veh_size>
		<perspective>view: left/right</perspective>
		<base_point>3D bbox base point(image): pt2</base_point>
		<vertex3d>3D bbox(world): pt1~pt8</vertex3d>
		<veh_loc_2d>3D bbox centroid(image)</veh_loc_2d>
	</object>
	<object>
            ...
	</object>
	...
</annotation>

Vehicle Type

Type	Annotation
car	Car
truck	Truck
bus	Bus

Download

All the images and annotations in SVLD-3D are available at Google Drive.

train/val set:

• DATA2021: Scene A-C
• DATA2021: Scene D
• DATA2021: Scene E

test set:

• TESTDATA2021: Scene A-E

Citation

@article{tang2022CenterLoc3D,
  title={CenterLoc3D: Monocular 3D Vehicle Localization Network for Roadside Surveillance Cameras},
  author={Tang, Xinyao and Song, Huansheng and Wang, Wei and Zhao, Chunhui},
  journal={arXiv preprint arXiv:2203.14550},
  year={2022}
}