Straw6D: In-Field 6D Pose Dataset for Robotic Strawberry Harvesting

This dataset accompanies the paper:

From Simulation to the Real-World: An In-Field 6D Pose Dataset and Baseline for Robotic Strawberry Harvesting Woojung Son, Won Suk Lee, Zijing Huang, Daeun Choi, Catia Silva, Yu She, Yan Gu University of Florida & Purdue University [arXiv]

Overview

To the best of our knowledge, Straw6D is the first real-world 6D pose ground truth dataset of strawberries collected in actual agricultural fields. Collecting 6D pose ground truth in real agricultural environments is inherently challenging; prior work has largely relied on synthetic data with limited realism. This dataset bridges that gap by providing per-frame 6D pose annotations paired with RGB images captured directly in the field.

• 12,040 RGB images with 16,037 annotated strawberry instances
• Single category: strawberry (red-stage / ripe only)
• Recorded using an Intel RealSense D435i camera

Annotation Pipeline

Ground truth 6D poses were obtained through a multi-step process:

1. Checkerboard-based camera calibration (PnP algorithm)
2. Metric-scale 3D reconstruction via COLMAP
3. Manual 3D bounding box annotation on reconstructed point clouds
4. 6D pose derived from camera-to-world and local-to-world transformations

Dataset Structure

Each sample provides image, per-frame camera parameters, and per-object pose annotations.

Camera fields (per frame):

Field	Description
image_id	Zero-padded frame index (e.g. 000042)
image	RGB image
resolution	[width, height] in pixels
fx, fy, cx, cy	Camera intrinsics in pixels
camera_view_matrix	4×4 world-to-camera transform
camera_projection_matrix	4×4 OpenGL projection matrix

Object fields (list, one entry per object in the frame):

Field	Description
labels	Category label ("strawberry")
bbox_2d_loose	2D bounding box [x_min, y_min, x_max, y_max] in pixels
size_local	Object size [x, y, z] in meters
center_local	Object center in local space
local_to_world_transform	4×4 object-local-to-world transform

All 4×4 matrices are row-major: p_out = p_in @ M.

Usage

from datasets import load_dataset

ds = load_dataset("WoojungSon/Straw6D")
sample = ds["train"][0]

print(sample["image_id"])   # "000000"
print(sample["labels"])     # ["strawberry"]
sample["image"]             # PIL Image

Visualization

A visualization script is provided in scripts/visualization.py. It projects the 6D pose axes onto the RGB image:

python scripts/visualization.py \
    --image path/to/rgb/000000.png \
    --json  path/to/json/000000.json \
    --output output.png

Citation

@article{son2025straw6d,
  title={From Simulation to the Real-World: An In-Field 6D Pose Dataset and Baseline for Robotic Strawberry Harvesting},
  author={Son, Woojung and Lee, Won Suk and Huang, Zijing and Choi, Daeun and Silva, Catia and She, Yu and Gu, Yan},
  journal={arXiv preprint arXiv:2606.11381},
  year={2025}
}

License

This dataset is released under the Creative Commons Attribution 4.0 International (CC BY 4.0) license.