MMPD
收藏资源简介:
MMPD数据集是由清华大学、商汤科技研究与Tetras.AI、香港大学和上海人工智能实验室共同创建的大规模多模态行人检测基准。该数据集涵盖了多种传感器模态,包括RGB、红外、深度、激光雷达和事件数据,以及多种模态组合。数据集的创建过程包括整合现有公共数据集和收集新的EventPed数据集,旨在解决复杂场景下行人检测的问题。MMPD数据集的应用领域广泛,包括自动驾驶、机器人和视频监控等,旨在通过多模态数据提高行人检测的准确性和鲁棒性。
The MMPD dataset is a large-scale multimodal pedestrian detection benchmark jointly developed by Tsinghua University, SenseTime Research, Tetras.AI, The University of Hong Kong, and the Shanghai AI Laboratory. It supports multiple sensor modalities including RGB, infrared, depth, LiDAR, and event data, alongside diverse modality combinations. The development of the MMPD dataset entails integrating existing public datasets and curating the novel EventPed dataset, aiming to address the core challenges of pedestrian detection in complex real-world scenarios. The dataset has broad application prospects in fields such as autonomous driving, robotics, and video surveillance, and is designed to improve the accuracy and robustness of pedestrian detection via multimodal data.
MMPedestron 数据集概述
数据集配置和模型
区域提议性能
预训练阶段
| 方法与配置 | 主干网络 | 下载链接 |
|---|---|---|
| MMPedestron | UNIXViT | Google Drive, Baidu Yun (Code: mmpd) |
CrowdHuman
| 方法与配置 | 主干网络 | 下载链接 |
|---|---|---|
| MMPedestron | UNIXViT | Google Drive, Baidu Yun (Code: mmpd) |
COCO-Person
| 方法与配置 | 主干网络 | 下载链接 |
|---|---|---|
| MMPedestron finetune | UNIXViT | Google Drive, Baidu Yun (Code: mmpd) |
FLIR
| 方法与配置 | 主干网络 | 下载链接 |
|---|---|---|
| MMPedestron | UNIXViT | Google Drive, Baidu Yun (Code: mmpd) |
PEDRo
| 方法与配置 | 主干网络 | 下载链接 |
|---|---|---|
| MMPedestron | UNIXViT | Google Drive, Baidu Yun (Code: mmpd) |
| MMPedestron(10% train data) | UNIXViT | Google Drive, Baidu Yun (Code: mmpd) |
LLVIP 数据集
| 方法与配置 | 主干网络 | 下载链接 |
|---|---|---|
| MMPedestron | UNIXViT | Google Drive, Baidu Yun (Code: mmpd) |
InOutDoor 数据集
| 方法与配置 | 主干网络 | 下载链接 |
|---|---|---|
| MMPedestron | UNIXViT | Google Drive, Baidu Yun (Code: mmpd) |
STCrowd 数据集
| 方法与配置 | 主干网络 | 下载链接 |
|---|---|---|
| MMPedestron | UNIXViT | Google Drive, Baidu Yun (Code: mmpd) |
EventPed 数据集
| 方法与配置 | 主干网络 | 下载链接 |
|---|---|---|
| MMPedestron | UNIXViT | Google Drive, Baidu Yun (Code: mmpd) |
融合实验
LLVIP
| 方法与配置 | 主干网络 | 下载链接 |
|---|---|---|
| MMPedestron | UNIXViT | Google Drive, Baidu Yun (Code: mmpd) |
InOutDoor
| 方法与配置 | 主干网络 | 下载链接 |
|---|---|---|
| MMPedestron | UNIXViT | Google Drive, Baidu Yun (Code: mmpd) |
STCrowd
| 方法与配置 | 主干网络 | 下载链接 |
|---|---|---|
| MMPedestron | UNIXViT | Google Drive, Baidu Yun (Code: mmpd) |
EventPed
| 方法与配置 | 主干网络 | 下载链接 |
|---|---|---|
| MMPedestron | UNIXViT | Google Drive, Baidu Yun (Code: mmpd) |
数据准备
请从以下链接获取数据集:MMPD-Dataset
训练与测试
训练
使用 Slurm 管理训练任务:
shell sh tools/slurm_train.sh ${PARTITION} ${JOB_NAME} ${CONFIG_FILE} ${WORK_DIR} ${GPUS}
测试
使用 Slurm 管理测试任务:
shell sh tools/slurm_test.sh ${PARTITION} ${JOB_NAME} ${CONFIG_FILE} ${CHECKPOINT} ${GPUS}
许可证
代码和数据可自由用于非商业用途,并可在这些条件下重新分发。对于商业查询,请联系 Mr. Sheng Jin (jinsheng13[at]foxmail[dot]com)。我们将向您发送详细协议。
引用
如果您发现我们的论文和代码对您的研究有用,请考虑给予星标和引用:
bibtex @inproceedings{zhang2024when, title={When Pedestrian Detection Meets Multi-Modal Learning: Generalist Model and Benchmark Dataset}, author={Zhang, Yi and Zeng, Wang and Jin, Sheng and Qian, Chen and Luo, Ping and Liu, Wentao}, booktitle={European Conference on Computer Vision (ECCV)}, year={2024}, month={September} }

- 1When Pedestrian Detection Meets Multi-Modal Learning: Generalist Model and Benchmark Dataset清华大学、商汤科技研究与Tetras.AI、香港大学、上海人工智能实验室 · 2024年



