The Semantic KITTI Dataset
Semantic-Kitti is a large semantic segmentation and scene understanding dataset developed for LiDAR-based autonomous driving. But what it is and what is it for?
Created on September 15|Last edited on October 3
Comment
What Is The Semantic KITTI Dataset?
The Semantic KITTI dataset contains annotated sequences of the KITTI Vision Odometry Benchmark and provides dense point-wise annotations for the complete 360-degree field-of-view of the employed automotive LiDAR.
Semantic KITTI is the largest dataset with sequential point-cloud information with over 23201 scans annotated in 28 classes.
What We're Covering About Semantic KITTI
What Is The Semantic KITTI Dataset?General Info About The Semantic KITTI DatasetSupported Tasks Of The Semantic KITTI DatasetRecommended Reading
General Info About The Semantic KITTI Dataset
Supported Tasks Of The Semantic KITTI Dataset
Here are the tasks supported by the Semantic KITTI dataset:
Semantic Segmentation
A model is required to output a label for each point of a scan, i.e., one full turn of the rotating LiDAR sensor. The Semantic KITTI dataset provides two settings for this task:
- With single scan, the model doesn’t need to distinguish between moving and non-moving objects, i.e., moving and non-moving are mapped to a single class.
- With multiple scans, however, moving and non-moving objects are delineated, which makes the task harder, since the method has to decide if something is dynamic.
Panoptic Segmentation
Panoptic segmentation is an extension of the semantic segmentation task but in addition to segmenting point clouds the model is also tasked with identifying the individual instances of thing classes. For example, identifying objects in a scene as well as labeling them.
The panoptic quality (PQ) proposed by Kirillov *et al.* can be used as a metric to evaluate models on this task.
Panoptic4D Segmentation
In 4D panoptic segmentation of point-cloud sequences, it is necessary to provide instance IDs and semantic labels for a sequence of scans. The instance ID needs to be uniquely assigned to each instance in both, space and time.
Again, here the class-wise instanceID is only required for thing classes that are "instantiable" and will be ignored for the stuff classes.
Moving Object Segmentation
Moving object segmentation of point cloud sequences requires distinguishing between dynamic and static objects in a scan. Commonly used metrics include Jaccard Index or intersection-over-union (mIoU) metric over moving and non-moving parts of the environment.
Semantic Scene Completion
In semantic scene completion, we are required to complete a scene inside a certain volume from a single initial scan.
More specifically, given an input voxel grid, where each voxel is marked as empty or occupied, depending on whether or not it contains a laser measurement one needs to predict whether a voxel is occupied and its semantic label in the completed scene.
Again Intersection-over-union can be used as a metric to evaluate the models trained for this task.
Recommended Reading
The nuScenes Dataset
nuScenes is a large-scale 3D perception dataset for Autonomous Driving provided by motional. The dataset has 3D bounding boxes for 1000 scenes.
The Woven Planet (Lyft) Level 5 Dataset
In this article, we'll be exploring the Woven Planet (Lyft) Level 5 dataset. We'll look at what it is as well as the autonomous vehicle tasks and techniques it supports
The Berkeley Deep Drive (BDD110K) Dataset
The BDD100K dataset is the largest and most diverse driving video dataset with 100,000 videos annotated for 10 different perception tasks in autonomous driving.
Object Detection for Autonomous Vehicles (A Step-by-Step Guide)
Digging into object detection and perception for autonomous vehicles using YOLOv5 and Weights & Biases
Add a comment
Iterate on AI agents and models faster. Try Weights & Biases today.