BennyLin01/virtual_tree

🌲 TreeScapes-3D: A Multimodal Dataset for Forest and Plant Scene Understanding

Welcome to the TreeScapes-3D dataset repository! This is a high-quality, multimodal 3D vision dataset specifically designed for complex natural plant environments (such as agriculture, forestry robotics navigation, and plant phenotyping). We provide exceptionally rich, perfectly spatially aligned annotations for every frame in the scene.

📸 Dataset Visualization

We have rendered extremely rich multimodal data for every viewpoint in the scene. Below is a real sample slice from the tree_001 scene:

Figure: RGB, Depth map, Disparity map, World Normal map, Instance Segmentation, and the corresponding 3D Point Cloud from the exact same viewpoint.

---

🚀 Data Modalities

As shown in the figure above, every frame in the dataset perfectly includes the following 6 fully aligned data modalities:

1. 📷 RGB
   • Showcases realistic lighting effects and plant textures, suitable for general object detection and image segmentation.
2. 📏 Depth (m)
   • High-precision absolute depth information, displayed as a heatmap (brighter colors indicate greater distance). Ideal for monocular/stereo depth estimation tasks.
3. 👁️ Disparity (px)
   • Disparity information corresponding to binocular stereo vision, providing perfect Ground Truth for stereo matching algorithms.
4. 🌐 World Normal [R=X, G=Y, B=Z]
   • Describes the 3D surface orientation of plant leaves and branches. This is a crucial feature for plant phenotyping and 3D reconstruction.
5. 🧩 Instance Segmentation
   • each object = distinct colour: Every single leaf and branch is assigned an independent and unique ID and color. This greatly advances research in instance segmentation of dense leaves with severe occlusions.
6. ☁️ Point Cloud (.ply)
   • High-precision point cloud data with real (X, Y, Z) coordinates (e.g., 30,000 pts per view). Ready to drag and rotate, suitable for 3D object detection and point cloud semantic segmentation.

---

📂 Dataset Structure

The structure of the downloaded dataset is as follows:

TreeScapes-3D/
├── train/
│   ├── tree_001/
│   │   ├── left_rgb/
│   │   ├── left_depth/
│   │   ├── left_disparity/
│   │   ├── left_normal/
│   │   ├── left_instance_seg/
│   │   └── point_clouds/       <-- contains .ply files
│   └── ...
├── val/
└── test/

💻 Quick Start

You can use the datasets library to quickly download and load this dataset:

from datasets import load_dataset

# Load the dataset
dataset = load_dataset("your_username/TreeScapes-3D")

# Check the first sample in the training set
sample = dataset['train'][0]
print(sample.keys()) # ['rgb', 'depth', 'normal', 'instance_seg', 'point_cloud_path']

# Display the RGB image
sample['rgb'].show()

📜 License & Citation

This project is open-sourced under the CC-BY 4.0 license. If you use this dataset in your research or project, please cite our work.

---

💡 Tip: If you have any questions, feel free to open a Discussion in the Community tab!