HuangZhe885 / RoCo

RoCo （MM2024 Oral）

Robust Cooperative Perception By Iterative Object Matching and Pose Adjustment

Installation

You can refer to OpenCOOD data introduction and OpenCOOD installation guide to prepare data and install RoCo. The installation is totally the same as CoAlign.

Data Preparation

mkdir a dataset folder under RoCo. Put your OPV2V, V2XSet, DAIR-V2X data in this folder. You just need to put in the dataset you want to use. RoCo/dataset. All data configurations are the same as CoAlign. For details, please refer to CoAlign.

├── my_dair_v2x 
│   ├── v2x_c
│   ├── v2x_i
│   └── v2x_v
├── OPV2V
│   ├── additional
│   ├── test
│   ├── train
│   └── validate
├── V2XSET
│   ├── test
│   ├── train
│   └── validate

Checkpoints and Results

Download them and save them to opencood/logs

How to use

We are improving our project platform based on CoAlign. You just need to replace the box_align_v2.py and intermedia_fusion_dataset.py files.
If you want to visualize the pose error, use evaluate_pose_graph.py in the tool folder.
Important: During the graph matching and optimization process, the parameter candidate_radius needs to be adjusted according to different datasets. For specific parameter details, refer to the experiments in RoCo.

RoCo/models/sub_modules/box_align_v2.py

Line 449 in bf9747b

candidate_radius = 2
The bounding boxes used in RoCo also come from saved files. You can download and save to opencood/logs,

Acknowlege

This project is impossible without the code of OpenCOOD, g2opy and d3d.

Thanks to @DerrickXuNu and @yifanlu0227 for the great code framework.

Once again, my sincere thanks to @yifanlu0227 for his patient and meticulous help.

Video

878_1721619587.mp4