Datasets:

ArXiv:
License:
BGLab commited on
Commit
a1ebfc2
·
verified ·
1 Parent(s): 2692018

Revise README.md

Browse files
Files changed (1) hide show
  1. README.md +23 -19
README.md CHANGED
@@ -2,25 +2,29 @@
2
  # AgriField3D: A Curated 3D Point Cloud Dataset of Field-Grown Plants From A Maize Diversity Panel
3
 
4
  ## Overview
5
- The application of artificial intelligence (AI) in three-dimensional (3D) agricultural research, particularly for maize,
6
- has been limited by the scarcity of large-scale, diverse datasets. While 2D image datasets are abundant, they cannot capture
7
- essential structural details such as leaf architecture, plant volume, and spatial arrangements that 3D data can provide.
8
- To address this limitation, we present a curated dataset of 3D point clouds of fully field-grown maize plants with diverse
9
- genetic backgrounds designed to be AI-ready for advancing agricultural research. Our dataset comprises over 1,000 high-quality
10
- point clouds representing diverse maize varieties collected using a Terrestrial Laser Scanner. To enhance the usability of the
11
- dataset, we performed graph-based segmentation to isolate individual leaves and stalk point clouds. Each leaf is assigned a
12
- consistent color label across all plants based on its order from bottom to top (e.g., all first leaves are the same color,
13
- all second leaves are the same color, and so on). Similarly, all stalks are assigned a single, distinct color. A rigorous
14
- quality control process was undertaken to manually correct any errors in the segmentation and leaf ordering. This ensures
15
- accurate segmentation and consistent color labeling and facilitates precise leaf counting and structural analysis. Additionally,
16
- the dataset includes metadata describing point cloud quality, the number of leaves, and the presence of tassels and maize cobs.
17
- To support diverse AI applications, we provide code that can sub-sample each point cloud to create a user-defined resolution
18
- (100k, 50k, 10k points, etc.) through uniform downsampling. All versions were manually quality-checked to ensure the preservation
19
- of topology and plant structure. This dataset lays the foundation for leveraging 3D data to enable advanced applications in agricultural
20
- research, particularly for maize phenotyping and plant structure studies.
21
- ## Dataset Structure
22
-
23
- The dataset consists of two compressed `.zip` files, which contain the 3D point cloud data:
 
 
 
 
24
 
25
  ```
26
  AgriField3D/
 
2
  # AgriField3D: A Curated 3D Point Cloud Dataset of Field-Grown Plants From A Maize Diversity Panel
3
 
4
  ## Overview
5
+ The use of artificial intelligence (AI) in three-dimensional (3D) agricultural research, especially for maize,
6
+ has been limited due to the lack of large-scale, diverse datasets. While 2D image datasets are widely available,
7
+ they fail to capture key structural details like leaf architecture, plant volume, and spatial arrangements—information
8
+ that 3D data can provide. To fill this gap, we present a carefully curated dataset of 3D point clouds representing fully
9
+ field-grown maize plants with diverse genetic backgrounds. This dataset is designed to be AI-ready, offering valuable
10
+ insights for advancing agricultural research.
11
+
12
+ Our dataset includes over 1,000 high-quality point clouds of maize plants, collected using a Terrestrial Laser Scanner.
13
+ These point clouds encompass various maize varieties, providing a comprehensive and diverse dataset. To enhance usability,
14
+ we applied graph-based segmentation to isolate individual leaves and stalks. Each leaf is consistently color-labeled based
15
+ on its position in the plant (e.g., all first leaves share the same color, all second leaves share another color, and so on).
16
+ Similarly, all stalks are assigned a unique, distinct color.
17
+
18
+ A rigorous quality control process was applied to manually correct any segmentation or leaf-ordering errors, ensuring
19
+ accurate segmentation and consistent labeling. This process facilitates precise leaf counting and structural analysis.
20
+ In addition, the dataset includes metadata describing point cloud quality, leaf count, and the presence of tassels and maize cobs.
21
+
22
+ To support a wide range of AI applications, we also provide code that allows users to sub-sample the point clouds,
23
+ creating versions with user-defined resolutions (e.g., 100k, 50k, 10k points) through uniform downsampling.
24
+ Every version of the dataset has been manually quality-checked to preserve plant topology and structure.
25
+ This dataset sets the stage for leveraging 3D data in advanced agricultural research, particularly for maize phenotyping and plant structure studies.
26
+
27
+ ## Dataset Directory Structure
28
 
29
  ```
30
  AgriField3D/