Dataset Summary

This dataset consist of 21,384 2448x2048 pairs of synthetic images for tomato plants. Each pair consist of left/right RGBD, and panoptic segmentation labels for the left image. See example_load.py for usage. sample loads a 835MB subset with 108 pairs. all loads the full 160GB dataset with 21,384 pairs.

Supported Tasks and Leaderboards

image-segmentation: Both panoptic and semantic labels for separating tomato plants and identifying features and disease types in the dataset.
depth-estimation: ground truth depth values for stereo and monocular applications.

Languages

English

Dataset Structure

Data Instances

Each datapoint consist of 6 images:

{
    'left_rgb': <PIL.JpegImagePlugin.JpegImageFile image mode=RGB size=2448x2048 at 0x7F63FB5F4350>, 
    'right_rgb': <PIL.JpegImagePlugin.JpegImageFile image mode=RGB size=2448x2048 at 0x7F63FF2B3950>, 
    'left_semantic': <PIL.PngImagePlugin.PngImageFile image mode=L size=2448x2048 at 0x7F63FC4488C0>, 
    'left_instance': <PIL.TiffImagePlugin.TiffImageFile image mode=I;16 size=2448x2048 at 0x7F63FC497EF0>, 
    'left_depth': <PIL.TiffImagePlugin.TiffImageFile image mode=F size=2448x2048 at 0x7F63FACF6E70>, 
    'right_depth': <PIL.TiffImagePlugin.TiffImageFile image mode=F size=2448x2048 at 0x7F63FACF7560>
}

Data Fields

'left_rgb': Left RGB image, was compressed to 95% quality.
'right_rgb': Right RGB image, was compressed to 95% quality. Note the baseline is 3.88112 cm and HFOV is 95.452621 degrees.
'left_semantic': Rendered colors that denotes the RGB label for individual pixels. See example_load.py for classes and sample scripts.
'left_instance': Rendered colors that denotes the tomato plant instances for individual pixels.
'left_depth': Rendered left depth compressed to 16-bit floats (in centimeters).
'right_depth': Rendered right depth compressed to 16-bit floats (in centimeters).

Data Splits

80/20 as shown in the train.txt and val.txt.

Dataset Creation

Curation Rationale

Created to provide dataset for dense plant disease detection for an agricultural research robotics platform with corresponding camera sensors and strobe lighting.

Source Data

Initial Data Collection and Normalization

We used PlantVillage Dataset with further processing to align the healthy leaf colors with the purchased assets. We collected 750GB of original data where we compressed the depth images from 32-bit to 16-bit and RGB to 95% quality for ~160GB.

Who are the source language producers?

See PlantVillage Datasets for tomato diseases. The tomato plants were purchased through SketchFab with modifications for extra green tomatoes and denser leaves.

Annotations

Annotation process

Annotations automatically generated through the textures in the simulation. The textures (two images) were labeled by dataset creators. The disease textures labels were labeled by PlantVillage creators that consist of experts in plant diseases.

Who are the annotators?

Same as dataset creators. Tomato leaf diseases the same as PlantVillage creators.

Personal and Sensitive Information

[More Information Needed]

Considerations for Using the Data

Social Impact of Dataset

[More Information Needed]

Discussion of Biases

[More Information Needed]

Other Known Limitations

[More Information Needed]

Additional Information

Dataset Curators

[More Information Needed]

Licensing Information

CC BY-NC-4.0

Citation Information

[More Information Needed]

Contributions

[More Information Needed]