Hugging Face's logo

Datasets:
Voxel51
/
crops3d

Tasks:
Image Classification
Modalities:
3D
Languages:
English
Size:
1K<n<10K
Tags:
3d
fiftyone
image-classification
Libraries:
FiftyOne
Dataset card Files
xet
Community
Dataset Viewer
The dataset viewer is not available for this split.
Cannot extract the features (columns) for the split 'train' of the config 'default' of the dataset.
Error code:   FeaturesError
Exception:    ArrowInvalid
Message:      Schema at index 1 was different: 
_id: struct<$oid: string>
name: string
slug: string
version: string
created_at: struct<$date: string>
last_modified_at: struct<$date: string>
last_deletion_at: null
last_loaded_at: struct<$date: string>
sample_collection_name: string
persistent: bool
media_type: string
group_media_types: struct<>
tags: list<item: null>
info: struct<>
app_config: struct<dynamic_groups_target_frame_rate: int64, grid_media_field: string, media_fallback: bool, media_fields: list<item: string>, modal_media_field: string, plugins: struct<>>
classes: struct<>
default_classes: list<item: null>
mask_targets: struct<>
default_mask_targets: struct<>
skeletons: struct<>
sample_fields: list<item: struct<name: string, ftype: string, embedded_doc_type: string, subfield: string, fields: list<item: struct<name: string, ftype: string, embedded_doc_type: null, subfield: string, fields: list<item: null>, db_field: string, description: null, info: null, read_only: bool, created_at: struct<$date: string>>>, db_field: string, description: null, info: null, read_only: bool, created_at: struct<$date: string>>>
frame_fields: list<item: null>
saved_views: list<item: null>
workspaces: list<item: null>
annotation_runs: struct<>
brain_methods: struct<>
evaluations: struct<>
runs: struct<>
vs
samples: list<item: struct<_id: struct<$oid: string>, filepath: string, tags: list<item: null>, _media_type: string, _rand: double, crop_type: struct<_id: struct<$oid: string>, _cls: string, tags: list<item: null>, label: string>, _dataset_id: struct<$oid: string>, created_at: struct<$date: string>, last_modified_at: struct<$date: string>>>
Traceback:    Traceback (most recent call last):
                File "/src/services/worker/src/worker/job_runners/split/first_rows.py", line 228, in compute_first_rows_from_streaming_response
                  iterable_dataset = iterable_dataset._resolve_features()
                File "/src/services/worker/.venv/lib/python3.9/site-packages/datasets/iterable_dataset.py", line 3422, in _resolve_features
                  features = _infer_features_from_batch(self.with_format(None)._head())
                File "/src/services/worker/.venv/lib/python3.9/site-packages/datasets/iterable_dataset.py", line 2187, in _head
                  return next(iter(self.iter(batch_size=n)))
                File "/src/services/worker/.venv/lib/python3.9/site-packages/datasets/iterable_dataset.py", line 2391, in iter
                  for key, example in iterator:
                File "/src/services/worker/.venv/lib/python3.9/site-packages/datasets/iterable_dataset.py", line 1882, in __iter__
                  for key, pa_table in self._iter_arrow():
                File "/src/services/worker/.venv/lib/python3.9/site-packages/datasets/iterable_dataset.py", line 1904, in _iter_arrow
                  yield from self.ex_iterable._iter_arrow()
                File "/src/services/worker/.venv/lib/python3.9/site-packages/datasets/iterable_dataset.py", line 559, in _iter_arrow
                  yield new_key, pa.Table.from_batches(chunks_buffer)
                File "pyarrow/table.pxi", line 4116, in pyarrow.lib.Table.from_batches
                File "pyarrow/error.pxi", line 154, in pyarrow.lib.pyarrow_internal_check_status
                File "pyarrow/error.pxi", line 91, in pyarrow.lib.check_status
              pyarrow.lib.ArrowInvalid: Schema at index 1 was different: 
              _id: struct<$oid: string>
              name: string
              slug: string
              version: string
              created_at: struct<$date: string>
              last_modified_at: struct<$date: string>
              last_deletion_at: null
              last_loaded_at: struct<$date: string>
              sample_collection_name: string
              persistent: bool
              media_type: string
              group_media_types: struct<>
              tags: list<item: null>
              info: struct<>
              app_config: struct<dynamic_groups_target_frame_rate: int64, grid_media_field: string, media_fallback: bool, media_fields: list<item: string>, modal_media_field: string, plugins: struct<>>
              classes: struct<>
              default_classes: list<item: null>
              mask_targets: struct<>
              default_mask_targets: struct<>
              skeletons: struct<>
              sample_fields: list<item: struct<name: string, ftype: string, embedded_doc_type: string, subfield: string, fields: list<item: struct<name: string, ftype: string, embedded_doc_type: null, subfield: string, fields: list<item: null>, db_field: string, description: null, info: null, read_only: bool, created_at: struct<$date: string>>>, db_field: string, description: null, info: null, read_only: bool, created_at: struct<$date: string>>>
              frame_fields: list<item: null>
              saved_views: list<item: null>
              workspaces: list<item: null>
              annotation_runs: struct<>
              brain_methods: struct<>
              evaluations: struct<>
              runs: struct<>
              vs
              samples: list<item: struct<_id: struct<$oid: string>, filepath: string, tags: list<item: null>, _media_type: string, _rand: double, crop_type: struct<_id: struct<$oid: string>, _cls: string, tags: list<item: null>, label: string>, _dataset_id: struct<$oid: string>, created_at: struct<$date: string>, last_modified_at: struct<$date: string>>>

Need help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.

Dataset Card for crops3d

image/png

This is a FiftyOne dataset with 1180 samples.

Installation

If you haven't already, install FiftyOne:

pip install -U fiftyone

Usage 

import json
import os


from huggingface_hub import snapshot_download


# Download the dataset snapshot to the current working directory

snapshot_download(repo_id="Voxel51/crops3d", local_dir=".", repo_type="dataset")

import fiftyone as fo

import fiftyone as fo
import os
import json

# Load dataset from current directory using FiftyOne's native format
dataset = fo.Dataset.from_dir(
    dataset_dir=".",  # Current directory contains the dataset files
    dataset_type=fo.types.FiftyOneDataset,  # Specify FiftyOne dataset format
    name="crops3d"  # Assign a name to the dataset for identification
)

def update_dataset_ply_paths(dataset):
    """
    Update PLY file paths in FiftyOne 3D dataset to use absolute paths.
    
    This function iterates through all samples in a FiftyOne dataset and modifies
    any PLY mesh file paths found in the JSON metadata to be absolute paths
    relative to the sample's directory location.
    
    Args:
        dataset (fiftyone.Dataset): A FiftyOne dataset containing 3D samples
                                   with JSON metadata files that may reference
                                   PLY mesh files with relative paths.
    
    Returns:
        None: The function modifies the JSON files in-place on disk.
    
    Note:
        - This function assumes each sample's filepath points to a JSON file
        - The JSON structure should contain a 'children' array with objects
        - PLY mesh objects are identified by '_type': 'PlyMesh'
        - Only objects with a 'plyPath' field will be updated
    """
    # Iterate through each sample in the dataset
    for sample in dataset:
        # Get the file path of the current sample (JSON metadata file)
        fo3d_filepath = sample.filepath
        
        # Extract the directory containing the sample file
        fo3d_directory = os.path.dirname(fo3d_filepath)
        
        # Read and parse the JSON metadata file
        with open(fo3d_filepath, 'r') as f:
            fo3d_data = json.load(f)
        
        # Process each child object in the JSON structure
        for child in fo3d_data.get('children', []):
            # Check if this child is a PLY mesh object with a path reference
            if child.get('_type') == 'PlyMesh' and 'plyPath' in child:
                # Convert relative PLY path to absolute path
                # by joining it with the sample's directory
                child['plyPath'] = os.path.join(fo3d_directory, child['plyPath'])
        
        # Write the updated JSON data back to the file
        with open(fo3d_filepath, 'w') as f:
            json.dump(fo3d_data, f, indent=2)  # Pretty-print with 2-space indentation

# Execute the path update function on the loaded dataset
update_dataset_ply_paths(dataset)

Dataset Details

Dataset Description

Curated by: Jianzhong Zhu, Ruifang Zhai, He Ren, Kai Xie, Aobo Du, Xinwei He, Chenxi Cui, Yinghua Wang, Junli Ye, Jiashi Wang, Xue Jiang, Yulong Wang, Chenglong Huang, Wanneng Yang (Huazhong Agricultural University, China)

Funded by: National Natural Science Foundation of China, National Key Research and Development Program of China

Shared by: Huazhong Agricultural University, National Key Laboratory of Crop Genetic Improvement

Language(s) (NLP): Not applicable (3D point cloud dataset)

License: Creative Commons Attribution 4.0 International (CC BY 4.0)

Dataset Sources [optional]

Repository: https://github.com/clawCa/Crops3D

Paper: https://doi.org/10.1038/s41597-024-04290-0

Demo: Available through FiftyOne visualization after processing with provided scripts

You can see how the dataset was parsed here

Uses

Direct Use

  • 3D crop phenotyping research: Analysis of plant morphology and structure

  • Instance segmentation: Segmenting individual plants in agricultural plots

  • Plant classification: Identifying and classifying different crop types from 3D data

  • Organ segmentation: Fine-grained segmentation of plant organs (leaves, stems, etc.)

  • Agricultural robotics: Training perception models for autonomous agricultural systems

  • Precision agriculture: Development of crop monitoring and assessment tools

  • Computer vision research: Benchmarking 3D segmentation and classification algorithms

Dataset Structure

The dataset contains 1,180 high-quality 3D point cloud samples in PLY format with RGB color information:

  • File format: PLY (Polygon File Format) with XYZ coordinates and RGB colors

  • Total samples: 1,180 point clouds

  • Crop types: 8 (Maize: 225, Cabbage: 196, Cotton: 176, Rapeseed: 150, Wheat: 148, Potato: 118, Rice: 84, Tomato: 83)

  • Acquisition methods: Multi-view stereo (mvs) and structured light (sl)

  • File naming: {CropType}-{identifier}.ply (e.g., Cabbage-mvs_1005_01.ply)

  • Point density: Varies by acquisition method and crop complexity

  • Additional variants:

    • Crops3D_10k: Subsampled to 10,000 points using Farthest Point Sampling
    • Crops3D_10k-C: Corruption robustness test sets with 7 corruption types at 5 severity levels
    • Crops3D_IS: Instance segmentation data for plot-level analysis

Dataset Creation

Curation Rationale

The dataset was created to address the lack of comprehensive 3D agricultural datasets for developing and benchmarking computer vision algorithms in precision agriculture. Existing datasets were limited in diversity, realism, and complexity, hindering the development of robust agricultural perception systems. Crops3D fills this gap by providing authentic, diverse, and complex 3D point cloud data from real agricultural environments.

Source Data

Data Collection and Processing

  • Collection period: Multiple growth stages across agricultural seasons

  • Location: Agricultural research fields in China

  • Equipment: Professional 3D scanning devices using multi-view stereo and structured light technologies

  • Processing:

    • Raw point clouds captured from multiple viewpoints
    • Registration and fusion to create complete 3D models
    • Color information preserved from RGB cameras
    • Quality control to ensure accurate geometry and color

Who are the source data producers?

The data was produced by researchers at:

  • Huazhong Agricultural University
  • National Key Laboratory of Crop Genetic Improvement
  • Collaborative teams from agricultural and computer vision research groups

Annotations

Annotation process

  • Plant-level labels: Each point cloud is labeled with its crop type
  • Instance segmentation: Plot-level data includes individual plant boundaries
  • Organ segmentation: Selected samples include organ-level annotations (leaves, stems)
  • Quality assurance: Multiple rounds of verification by agricultural experts

Who are the annotators?

  • Agricultural researchers and graduate students from Huazhong Agricultural University
  • Domain experts in crop phenotyping and plant science
  • Computer vision researchers familiar with 3D data annotation

Personal and Sensitive Information

The dataset contains no personal or sensitive information. All data consists of 3D scans of agricultural crops in research fields with no human subjects or private information included.

Bias, Risks, and Limitations

Biases:

  • Geographic bias: Data collected primarily from agricultural fields in China, may not represent all global agricultural conditions
  • Crop selection bias: Limited to 8 major crop types, doesn't cover all agricultural species
  • Growth stage bias: May not equally represent all growth stages for each crop

Risks:

  • Generalization: Models trained on this data may not perform well on crops grown in significantly different climates or conditions
  • Scale limitations: Individual plant models may not scale directly to large field applications

Limitations:

  • Point cloud density variations: Different acquisition methods produce varying point densities
  • Occlusion: Natural self-occlusion in dense crop canopies may limit visibility of internal structures
  • Weather conditions: Data collected under specific weather conditions may not represent all scenarios
  • File size: Large PLY files (~9GB total) require significant storage and processing resources

Recommendations

Users should be made aware of the following recommendations:

  1. Preprocessing: Consider subsampling large point clouds for efficiency (Crops3D_10k variant available)
  2. Validation: Test model generalization on crops from different geographic regions
  3. Augmentation: Apply appropriate 3D augmentations to improve model robustness
  4. Multi-modal approaches: Consider combining with 2D imagery or other sensor data for comprehensive analysis
  5. Computational resources: Ensure adequate GPU memory and storage for processing large point clouds
  6. Domain adaptation: May require fine-tuning for specific agricultural environments or crop varieties

Citation

BibTeX:

@article{zhu2024crops3d,
  title={Crops3D: a diverse 3D crop dataset for realistic perception and segmentation toward agricultural applications},
  author={Zhu, Jianzhong and Zhai, Ruifang and Ren, He and Xie, Kai and Du, Aobo and He, Xinwei and Cui, Chenxi and Wang, Yinghua and Ye, Junli and Wang, Jiashi and Jiang, Xue and Wang, Yulong and Huang, Chenglong and Yang, Wanneng},
  journal={Scientific Data},
  volume={11},
  number={1438},
  year={2024},
  doi={10.1038/s41597-024-04290-0},
  publisher={Nature Publishing Group}
}

APA:

Zhu, J., Zhai, R., Ren, H., Xie, K., Du, A., He, X., Cui, C., Wang, Y., Ye, J., Wang, J., Jiang, X., Wang, Y., Huang, C., & Yang, W. (2024). Crops3D: a diverse 3D crop dataset for realistic perception and segmentation toward agricultural applications. Scientific Data, 11, 1438. https://doi.org/10.1038/s41597-024-04290-0

Glossary

  • PLY: Polygon File Format, a file format for storing 3D data
  • Point Cloud: A collection of 3D points representing the external surface of an object
  • FPS: Farthest Point Sampling, a method for subsampling point clouds
  • Instance Segmentation: Identifying and separating individual objects (plants) in a scene
  • Organ Segmentation: Identifying different parts of a plant (leaves, stems, fruits)
  • Multi-view Stereo (MVS): 3D reconstruction technique using multiple 2D images
  • Structured Light (SL): 3D scanning method using projected light patterns
  • Phenotyping: Measuring observable plant characteristics

More Information

Dataset Card Authors

  • Original dataset: Jianzhong Zhu et al., Huazhong Agricultural University
  • Dataset card: Prepared for Hugging Face distribution

Dataset Card Contact

Downloads last month
688