WHALES
收藏资源简介:
WHALES数据集由清华大学开发,使用CARLA模拟器生成,专注于自动驾驶中的多智能体协同感知。数据集包含70K RGB图像、17K LiDAR帧和2.01M 3D边界框注释,涵盖多种道路场景。创建过程中,通过优化模拟速度和计算成本,实现了每驾驶序列平均8.4个智能体的记录。该数据集主要用于解决自动驾驶中的遮挡和感知范围有限的问题,支持V2V和V2I感知任务,推动合作感知技术的发展。
The WHALES dataset was developed by Tsinghua University and generated using the CARLA simulator, focusing on multi-agent collaborative perception in autonomous driving. It contains 70K RGB images, 17K LiDAR frames, and 2.01 million 3D bounding box annotations, covering a wide range of road scenarios. During its development, the simulation speed and computational cost were optimized, achieving an average of 8.4 agents per driving sequence. This dataset is primarily designed to address the challenges of occlusion and limited perception range in autonomous driving, supporting V2V and V2I perception tasks and advancing the development of cooperative perception technologies.
WHALES 数据集概述
数据集简介
WHALES(Wireless Enhanced Autonomous vehicles with Large number of Engaged agents)是一个由CARLA模拟器生成的自动驾驶数据集,旨在解决单系统感知范围有限和遮挡问题。该数据集平均每驾驶序列包含8.4个代理,提供了大规模的代理和视角,并记录了代理行为,支持多任务合作感知。
数据集特点
- 代理数量:平均每驾驶序列包含8.4个代理,是目前自动驾驶数据集中代理数量最多的。
- 任务支持:支持多任务合作感知,包括感知和规划任务。
- 传感器配置:包括LiDAR和摄像头,提供丰富的感知数据。
数据集比较
| 数据集 | 年份 | 真实/模拟 | V2X | 图像 | 点云 | 3D标注 | 类别 | 平均代理数 |
|---|---|---|---|---|---|---|---|---|
| KITTI | 2012 | 真实 | 否 | 15k | 15k | 200k | 8 | 1 |
| nuScenes | 2019 | 真实 | 否 | 1.4M | 400k | 1.4M | 23 | 1 |
| DAIR-V2X | 2021 | 真实 | V2V&I | 39k | 39k | 464k | 10 | 2 |
| V2X-Sim | 2021 | 模拟 | V2V&I | 0 | 10k | 26.6k | 2 | 2 |
| OPV2V | 2022 | 模拟 | V2V | 44k | 11k | 230k | 1 | 3 |
| DOLPHINS | 2022 | 模拟 | V2V&I | 42k | 42k | 293k | 3 | 3 |
| V2V4Real | 2023 | 真实 | V2V | 40k | 20k | 240k | 5 | 2 |
| WHALES (Ours) | 2024 | 模拟 | V2V&I | 70k | 17k | 2.01M | 3 | 8.4 |
代理类别
| 代理位置 | 代理类别 | 传感器配置 | 规划与控制 | 任务 | 生成位置 |
|---|---|---|---|---|---|
| 在路上 | 不受控CAV | LiDAR × 1 + 摄像头 × 4 | CARLA自动驾驶 | 感知 | 随机,确定性 |
| 在路上 | 受控CAV | LiDAR × 1 + 摄像头 × 4 | RL算法 | 感知与规划 | 随机,确定性 |
| 路边 | RSU | LiDAR × 1 + 摄像头 × 4 | RL算法 | 感知与规划 | 静态 |
| 任意位置 | 障碍物代理 | 无传感器 | CARLA自动驾驶 | 无任务 | 随机 |
实验结果
单系统3D目标检测基准(50m/100m)
| 方法 | $ ext{AP}_{Veh}uparrow$ | $ ext{AP}_{Ped}uparrow$ | $ ext{AP}_{Cyc}uparrow$ | $mAPuparrow$ | $mATEdownarrow$ | $mASEdownarrow$ | $mAOEdownarrow$ | $mAVEdownarrow$ | $NDSuparrow$ |
|---|---|---|---|---|---|---|---|---|---|
| Pointpillars | 67.1/41.5 | 38.0/6.3 | 37.3/11.6 | 47.5/19.8 | 0.117/0.247 | 0.876/0.880 | 1.069/1.126 | 1.260/1.625 | 33.8/18.6 |
| SECOND | 58.5/38.8 | 27.1/12.1 | 24.1/12.9 | 36.6/21.2 | 0.106/0.156 | 0.875/0.878 | 1.748/1.729 | 1.005/1.256 | 28.5/20.3 |
| RegNet | 66.9/42.3 | 38.7/8.4 | 32.9/11.7 | 46.2/20.8 | 0.119/0.240 | 0.874/0.881 | 1.079/1.158 | 1.231/1.421 | 33.2/19.2 |
| VoxelNeXt | 64.7/42.3 | 52.2/27.4 | 35.9/9.0 | 50.9/26.2 | 0.075/0.142 | 0.877/0.877 | 1.212/1.147 | 1.133/1.348 | 36.0/22.9 |
合作3D目标检测基准(50m/100m)
| 方法 | $ ext{AP}_{Veh}uparrow$ | $ ext{AP}_{Ped}uparrow$ | $ ext{AP}_{Cyc}uparrow$ | $mAPuparrow$ | $mATEdownarrow$ | $mASEdownarrow$ | $mAOEdownarrow$ | $mAVEdownarrow$ | $NDSuparrow$ |
|---|---|---|---|---|---|---|---|---|---|
| No Fusion | 67.1/41.5 | 38.0/6.3 | 37.3/11.6 | 47.5/19.8 | 0.117/0.247 | 0.876/0.880 | 1.069/1.126 | 1.260/1.625 | 33.8/18.6 |
| F-Cooper | 75.4/52.8 | 50.1/9.1 | 44.7/20.4 | 56.8/27.4 | 0.117/0.205 | 0.874/0.879 | 1.074/1.206 | 1.358/1.449 | 38.5/22.9 |
| Raw-level Fusion | 71.3/48.9 | 38.1/8.5 | 40.7/16.3 | 50.0/24.6 | 0.135/0.242 | 0.875/0.882 | 1.062/1.242 | 1.308/1.469 | 34.9/21.1 |
| *VoxelNeXt | 71.5/50.6 | 60.1/35.4 | 47.6/21.9 | 59.7/35.9 | 0.085/0.159 | 0.877/0.878 | 1.070/1.204 | 1.262/1.463 | 40.2/27.6 |
不同调度策略下的mAP分数(50m/100m)
| 推理训练 | No Fusion | Closest Agent | Single Random | Multiple Random | Full Communication |
|---|---|---|---|---|---|
| No Fusion | 50.9/26.2 | 50.9/23.3 | 51.3/25.3 | 50.3/22.9 | 45.6/18.8 |
| Closest Agent | 39.9/20.3 | 58.4/30.2 | 58.3/32.6 | 57.7/30.5 | 55.4/10.8 |
| Single Random | 43.3/22.8 | 57.9/31.0 | 58.4/33.3 | 57.7/31.4 | 55.0/14.6 |
| MASS | 55.5/11.0 | 58.8/33.7 | 58.9/34.0 | 57.3/32.3 | 54.1/27.4 |
| Historical Best | 54.8/29.6 | 58.6/31.7 | 58.9/34.0 | 58.3/32.6 | 54.1/27.4 |
| Multiple Random | 34.5/16.9 | 60.7/35.1 | 61.2/37.1 | 61.4/36.4 | 58.8/12.9 |
| Full Communication | 29.1/10.5 | 63.7/38.4 | 64.0/39.9 | 64.7/41.3 | 65.1/39.2 |

- 1WHALES: A Multi-agent Scheduling Dataset for Enhanced Cooperation in Autonomous Driving清华大学 · 2024年



