CVPR 2023 Occupancy Prediction Challenge

The world's First 3D Occupancy Benchmark for Scene Perception in Autonomous Driving.

English | 中文

Table of Contents

• CVPR 2023 Occupancy Prediction Challenge
  • Table of Contents
  • Introduction
  • Task Definition
  • Evaluation Metrics
  • Timeline
  • Data
  • Development Kit
  • Leaderboard
  • License

Introduction

Understanding the 3D surroundings including the background stuffs and foreground objects is important for autonomous driving. In the traditional 3D object detection task, a foreground object is represented by the 3D bounding box. However, the geometrical shape of the object is complex, which can not be represented by a simple 3D box, and the perception of the background is absent. The goal of this task is to predict the 3D occupancy of the scene. In this task, we provide a large-scale occupancy benchmark based on the nuScenes dataset. The benchmark is a voxelized representation of the 3D space, and the occupancy state and semantics of the voxel in 3D space are jointly estimated in this task. The complexity of this task lies in the dense prediction of 3D space given the surround-view image.

(back to top)

Task Definition

Given images from multiple cameras, the goal is to predict the current occupancy state and semantics of each voxel grid in the scene. The voxel state is predicted to be either free or occupied. If a voxel is occupied, its semantic class needs to be predicted, as well. Besides, we also provide a binary known/unknown mask for each scene. An unknown voxel is defined as an invisible grid in the current camera observation, which is ignored in the evaluation stage.

Evaluation Metrics

Leaderboard ranking for this challenge is by the intersection-over-union (mIoU) over all classes.

mIoU

Let $C$ be he number of classes.

$$ mIoU=\frac{1}{C}\displaystyle \sum_{c=1}^{C}\frac{TP_c}{TP_c+FP_c+FN_c}, $$

where $TP_c$ , $FP_c$ , and $FN_c$ correspond to the number of true positive, false positive, and false negative predictions for class $c_i$.

F1 Score

We also measure the F-score as the harmonic mean of the completeness $P_c$ and the accuracy $P_a$.

$$ F-score=\left( \frac{P_a^{-1}+P_c^{-1}}{2} \right) ^{-1} , $$

where $P_a$ is the percentage of predicted voxels that are within a distance threshold to the ground truth voxels, and $P_c$ is the percentage of ground truth voxels that are within a distance threshold to the predicted voxels.

(back to top)

Timeline

• Feb 20, 2023 - Dataset and Devkit Release.
• Mar 21, 2023 - Challenge Period Open.
• Jun 01, 2023 - Challenge Period End.
• Jun 03, 2023 - Finalist Notification.
• Jun 10, 2023 - Technical Report Deadline.
• Jun 12, 2023 - Winner Announcement.

* All due at 23:59 UTC+8.

(back to top)

Data

To be released.

(back to top)

Development Kit

To be released.

(back to top)

Leaderboard

To be released.

(back to top)

License

Before using the dataset, you should register on the website and agree to the terms of use of the nuScenes. All code within this repository is under Apache License 2.0.

(back to top)