K-Radar is the world’s largest publicly available 4D radar dataset for autonomous driving, with the highest number of labeled frames among existing datasets. It is also the world’s first to provide full RAED (Range–Azimuth–Elevation–Doppler) tensor data without any loss of dimensionality. Among 4D radar datasets, K-Radar features the most diverse adverse weather scenarios, making it a uniquely comprehensive resource for robust perception research.
Comparison of characteristics of existing 4D radar datasets is as follows:
-
Most datasets provide point cloud formats that deliver a sparse representation after pre-processing of raw radar tensors data. This sparsity leads to degraded 3D object detection performance, which is why recent works are shifting towards utilizing tensor data that has high density [1–6].
-
Even when 4D radar data is collected in tensor format, released datasets include only partial RAED dimensions (2D or 3D tensor) [2, 7, 8].
-
To date, K-Radar [1] and RaDelft [7] provide tensor data across all RAED dimensions (4D). However, RaDelft has relatively low elevation resolution, resulting in unreliable height information, has limited utilization due to the insufficient number of data frames with labels [7]. In addition, K-Radar provides the largest number of data frames with labels among 4D radar dataset currently available [2, 8].
It is assumed that most 4D radar hardware providers restricted external release of raw tensor data due to technology protection and security policies, which made it difficult to open a complete 4D tensor dataset to public. On the contrary, K-Radar overcame these technical and policy barriers by publicly releasing full 4D tensor data for the first time, enabling high-quality training for more accurate and robust object detection and tracking.
Note:
Azi. = Azimuth, Ele. = Elevation, Velo. = Velocity (Doppler), Odo. = Odometry, N/M = Not Mentioned
RPC = Radar Point Cloud. Data formats: R/A/E/D = Range / Azimuth / Elevation / Doppler.
Sensor modalities: R = Radar, C = Camera, L = LiDAR, I = IMU, T = Thermal Camera. Tensor datasets are indicated with underlines.
Point Cloud: Astyx [9], VoD [13], TJ4DRadSet [14], NTU4DRadLM [15], Dual Radar [16], MSC-RAD4R [17]
3D Tensor: SCORP [10], ColoRadar [11], RADIaL [12], RaDelft (Effectively 3D) [7]
4D Tensor: K-Radar [1]
- Datasets with low elevation resolution are considered Effectively 3D, even if all RAED dimensions are included, as there is high uncertainty in the provided height information.
- With a total of 35K labeled frames, it is the largest number of labeled frames among publicly available 4D radar datasets that are sufficient for various AI training.
- K-Radar provides full RAED tensors (~300 MB per frame and a total size of ~16 TB) and the largest number of labeled frames among 4D radar datasets — known as the largest 4D radar dataset among the publicly available datasets.
- Among 4D radar datasets, K-Radar includes the most diverse adverse weather scenarios [8].
[1] Paek, Dong-Hee, Seung-Hyun Kong, and Kevin Tirta Wijaya. "K-radar: 4d radar object detection for autonomous driving in various weather conditions." Advances in Neural Information Processing Systems 35 (2022): 3819-3829.
[2] Kong, Seung-Hyun, Dong-Hee Paek, and Sangyeong Lee. "Rtnh+: Enhanced 4d radar object detection network using two-level preprocessing and vertical encoding." IEEE Transactions on Intelligent Vehicles (2024).
[3] Ding, Fangqiang, et al. "Radarocc: Robust 3d occupancy prediction with 4d imaging radar." Advances in Neural Information Processing Systems 37 (2024): 101589-101617.
[4] Cheng, Jen-Hao, et al. "Centerradarnet: Joint 3d object detection and tracking framework using 4d fmcw radar." 2024 IEEE International Conference on Image Processing (ICIP). IEEE, 2024.
[5] Fent, Felix, Andras Palffy, and Holger Caesar. "Dpft: Dual perspective fusion transformer for camera-radar-based object detection." IEEE Transactions on Intelligent Vehicles (2024).
[6] Liu, Yang, et al. "Echoes beyond points: Unleashing the power of raw radar data in multi-modality fusion." Advances in Neural Information Processing Systems 36 (2023): 53964-53982.
[7] Roldan, Ignacio, et al. "A deep automotive radar detector using the RaDelft dataset." IEEE Transactions on Radar Systems (2024).
[8] Peng, Xiangyuan, et al. "4D mmWave Radar in Adverse Environments for Autonomous Driving: A Survey." arXiv e-prints (2025): arXiv-2503.
[9] Meyer, Michael, and Georg Kuschk. "Automotive radar dataset for deep learning based 3d object detection." 2019 16th european radar conference (EuRAD). IEEE, 2019.
[10] Nowruzi, Farzan Erlik, et al. "Deep open space segmentation using automotive radar." 2020 IEEE MTT-S International Conference on Microwaves for Intelligent Mobility (ICMIM). IEEE, 2020.
[11] Kramer, Andrew, et al. "Coloradar: The direct 3d millimeter wave radar dataset." The International Journal of Robotics Research 41.4 (2022): 351-360.
[12] Rebut, Julien, et al. "Raw high-definition radar for multi-task learning." Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 2022.
[13] Palffy, Andras, et al. "Multi-class road user detection with 3+ 1d radar in the view-of-delft dataset." IEEE Robotics and Automation Letters 7.2 (2022): 4961-4968.
[14] Zheng, Lianqing, et al. "TJ4DRadSet: A 4D radar dataset for autonomous driving." 2022 IEEE 25th International Conference on Intelligent Transportation Systems (ITSC). IEEE, 2022.
[15] Zhang, Jun, et al. "Ntu4dradlm: 4d radar-centric multi-modal dataset for localization and mapping." 2023 IEEE 26th International Conference on Intelligent Transportation Systems (ITSC). IEEE, 2023.
[16] Zhang, Xinyu, et al. "Dual radar: A multi-modal dataset with dual 4d radar for autononous driving." Scientific data 12.1 (2025): 439.
[17] Choi, Minseong, et al. "Msc-rad4r: Ros-based automotive dataset with 4d radar." IEEE Robotics and Automation Letters 8.11 (2023): 7194-7201.