---
license: cc-by-4.0
pretty_name: 3D-CovDiffusion Train-Ready Dataset
tags:
  - robotics
  - robot-learning
  - industrial-robotics
  - 3d
  - point-cloud
  - diffusion-policy
  - trajectory-generation
  - coverage-path-planning
  - robotic-spray-painting
  - iros-2026
size_categories:
  - 1K<n<10K
---

# 3D-CovDiffusion Train-Ready Dataset

Processed artifacts for
[3D-CovDiffusion](https://github.com/crystalccy1/3D-CovDiffusion): compact
training tensors and self-contained evaluation inputs for the complete fixed
test splits. Users do not need to rebuild point clouds, trajectories, or mesh
geometry from raw records.

- Paper: https://arxiv.org/abs/2510.03011
- Project page: https://crystalccy1.github.io/3D-CovDiffusion/
- Code: https://github.com/crystalccy1/3D-CovDiffusion
- Models: https://huggingface.co/ChenyuanC/3D-CovDiffusion

The paper reports a jointly trained, category-agnostic model. The current
public code-and-model release provides separately trained checkpoints for
Windows, Cuboids, Shelves, and Containers; this dataset contains the matching
per-category training and fixed-test records.

## Contents

| Category | Training episodes | Training steps | Evaluation cases |
|:--|--:|--:|--:|
| Windows | 800 | 161,436 | 200 |
| Cuboids | 800 | 486,169 | 200 |
| Shelves | 800 | 459,439 | 200 |
| Containers | 70 | 37,175 | 18 |
| **Total** | **2,470** | **1,144,219** | **618** |

```text
3D-CovDiffusion-Train-Ready/
├── dataset_manifest.json
├── evaluation_cache_manifest.json
├── data/
│   ├── windows-v2/{manifest.json,train.zarr/}
│   ├── cuboids-v2/{manifest.json,train.zarr/}
│   ├── shelves-v2/{manifest.json,train.zarr/}
│   └── containers-v2/{manifest.json,train.zarr/}
└── evaluation-cache/
    ├── windows-v2/*.npz
    ├── cuboids-v2/*.npz
    ├── shelves-v2/*.npz
    └── containers-v2/*.npz
```

## Training schema

Each `train.zarr` uses schema `3dcov-train-v1`:

```text
train.zarr/
├── meta/episode_ends       int64   [episodes]
├── data/action             float32 [steps, 24]
├── data/stroke_ids         float32 [steps]
└── obs/point_cloud         float32 [episodes, 5120, 3]
```

Every 24-D action token concatenates four ordered 6-DoF poses. The model trains
on horizon-16 token sequences. Each static 5,120-point observation is stored
once per episode.

## Evaluation schema

The `3dcov-evaluation-ready-v2` manifest stores the ordered fixed-test sample
IDs and a SHA-256 digest for every NPZ. Each NPZ uses schema
`3dcov-evaluation-v2` and contains:

```text
schema_version  scalar string
sample_id       scalar string
point_cloud     float [5120, 3]
trajectory      float [T, 24]
gt_trajectory   float [P, 6]
stroke_ids      int   [P]
mesh_vertices   float [V, 3]
mesh_faces      int   [F, 3]
```

These 618 files contain the complete model and metric inputs. The released
loader performs validation only; it does not resample meshes or rebuild action
tokens at runtime.

## Download and use

The public code pins an immutable dataset revision:

```bash
git clone https://github.com/crystalccy1/3D-CovDiffusion.git
cd 3D-CovDiffusion
python -m pip install -r requirements.txt

python reproduce.py prepare windows --data-only
python reproduce.py prepare windows --checkpoint-only
python reproduce.py evaluate windows
```

Use `cuboids`, `shelves`, or `containers` for another category.

## Citation

```bibtex
@misc{chen2026_3dcovdiffusion,
  title  = {{3D-CovDiffusion}: 3D-Aware Diffusion Policy
            for Coverage Path Planning},
  author = {Chen, Chenyuan and Ding, Haoran and Ding, Ran
            and Liu, Tianyu and He, Zewen and Duan, Anqing
            and Nakamura, Yoshihiko},
  year   = {2026},
  eprint = {2510.03011},
  archivePrefix = {arXiv},
  primaryClass  = {cs.RO},
  url    = {https://arxiv.org/abs/2510.03011},
  note   = {Accepted at IROS 2026}
}
```

## Attribution and license

This work uses modified data derived from “MaskPlanner: Learning-Based
Object-Centric Motion Generation from 3D Point Clouds” by Gabriele Tiboni,
[Zenodo record 14967945](https://zenodo.org/records/14967945), licensed under
[CC BY 4.0](https://creativecommons.org/licenses/by/4.0/). The data were
converted and preprocessed by the 3D-CovDiffusion authors; no endorsement by
the original creator is implied.