Voxel51/CitrusFarm-Sample

Dataset Card for CitrusFarm sample (FiftyOne multimodal MCAP)

This is a FiftyOne dataset with 1 sample. The sample is one sequence of a ground robot driving through a citrus farm, stored as a native multimodal MCAP episode.

The source is CitrusFarm, the multimodal agricultural robotics dataset from the ARCS Lab at the University of California Riverside, converted from ROS 1 bags. A Clearpath Jackal drives the rows of citrus trees at the university's Agricultural Experimental Station carrying a FLIR Blackfly monochrome camera, a FLIR ADK thermal camera, a Mapir Survey3 camera that sees red, green and near-infrared, a Stereolabs ZED 2i stereo camera with its depth, a Velodyne LiDAR, a Microstrain inertial unit and a Piksi GPS-RTK receiver, and records its own wheel odometry. The release holds seven sequences from three fields, 1.3 TB in all; this sample is the smallest of them, 06_14B_Jackal, from field 14B.

Installation

If you haven't already, install FiftyOne:

pip install -U fiftyone

Usage

import fiftyone as fo
import fiftyone.utils.huggingface as fouh

# Load the dataset
# Note: other available arguments include 'max_samples', etc
dataset = fouh.load_from_hub(
    "Voxel51/CitrusFarm-Sample",
    name="CitrusFarm-Sample",
    persistent=True,
)

# Launch the App
session = fo.launch_app(dataset)

Dataset Details

Dataset Description

One sequence, 5m 03s of driving over 357 m, with 3,026 monochrome, 3,024 thermal, 3,025 red-green-NIR and 3,022 ZED frames, 2,998 LiDAR scans and 3,024 GPS-RTK fixes.

• Curated by: ARCS Lab, University of California Riverside (source release)
• Funded by: [More Information Needed]
• Shared by: Voxel51 (FiftyOne conversion)
• Language(s): Not applicable (sensor data)
• License: CC-BY-SA-4.0

Dataset Sources

• Repository: CitrusFarm project page (source release); Voxel51/CitrusFarm-Sample (this conversion)
• Paper: Multimodal Dataset for Localization, Mapping and Crop Monitoring in Citrus Tree Farms (International Symposium on Visual Computing, 2023)
• Demo: [More Information Needed]

Uses

Direct Use

Localization, mapping and crop monitoring in citrus tree farms, using the thermal, multispectral, stereo, depth, LiDAR, inertial and GPS-RTK streams recorded on one shared clock, with the release's ground-truth trajectory as a position reference.

Out-of-Scope Use

[More Information Needed]

Dataset Structure

Topology

An ungrouped FiftyOne dataset with media_type="multimodal". One sample is one sequence, and its filepath is a .fo.mcap file (data/06_14B_Jackal.fo.mcap). There is 1 sample, no sample tags, and an empty dataset.info. The dataset has no FiftyOne label fields; every signal lives inside the MCAP file as a channel and is shown by the App's multimodal viewer.

Sample fields

Field	FiftyOne type	Description
id, filepath, tags, metadata, created_at, last_modified_at	built-in	Standard FiftyOne sample fields
sequence	StringField	Sequence name, 06_14B_Jackal
field	StringField	The farm field the sequence was recorded in, 14B
recorded	StringField	Recording date, 2023-07-18
duration	FloatField	Sequence length in seconds
num_mono_frames	IntField	Frames from the Blackfly monochrome camera
num_thermal_frames	IntField	Frames from the ADK thermal camera
num_rgn_frames	IntField	Frames from the Survey3 red-green-NIR camera
num_zed_frames	IntField	Frames from the ZED color images
num_depth_frames	IntField	Frames of ZED depth
num_lidar_scans	IntField	Velodyne LiDAR scans
num_lidar_points	IntField	Points across all LiDAR scans
num_gnss_fixes	IntField	GPS-RTK fixes
valid_depth_fraction	FloatField	Share of ZED depth pixels holding a value
ground_truth_path_m	FloatField	Length of the ground-truth trajectory in metres

Sample values

Field	Value
sequence	06_14B_Jackal
field	14B
recorded	2023-07-18
duration	302.535 s
num_mono_frames	3,026
num_thermal_frames	3,024
num_rgn_frames	3,025
num_zed_frames	3,022
num_depth_frames	3,023
num_lidar_scans	2,998
num_lidar_points	76,519,363
num_gnss_fixes	3,024
valid_depth_fraction	0.4703
ground_truth_path_m	356.6

Episode contents

Each MCAP episode contains these channels:

Channel	Schema or content
/mono	Blackfly monochrome camera at 1440x1080, foxglove.CompressedVideo
/thermal	ADK thermal camera at 640x512, foxglove.CompressedVideo
/rgn	Survey3 camera's red, green and near-infrared channels at 1280x720, foxglove.CompressedVideo
/zed-left, /zed-right	The ZED's rectified color images at 1280x720, foxglove.CompressedVideo
/zed-depth	The ZED's depth registered to its left image, foxglove.CompressedImage (16-bit PNG, millimetres, 0 where the ZED gives no depth)
<camera>-calibration	A calibration topic beside each camera, foxglove.CameraCalibration
/lidar-points	Velodyne scans, foxglove.PointCloud with x, y, z, intensity and ring
/imu.plot	Microstrain inertial unit
/gnss	GPS-RTK fixes, foxglove.LocationFix
/wheel-odometry, /zed-odometry	The robot's and the ZED's own pose estimates, foxglove.PoseInFrame, each with its position on a .plot channel
/ground-truth.plot	The release's ground-truth trajectory
/tf	The sensors' poses from the release's calibration, foxglove.FrameTransform
/sequence	Names the sequence and its field

The episode also carries the metadata fields sequence, field, recorded, duration, num_mono_frames, num_thermal_frames, num_rgn_frames, num_zed_frames, num_depth_frames, num_lidar_scans, num_lidar_points, num_gnss_fixes, valid_depth_fraction and ground_truth_path_m, which are the sample fields above.

Parsing decisions

• Why MCAP: the dataset is one multimodal sample per sequence, so the five cameras, LiDAR, inertial unit, GPS-RTK fixes and poses play back on one shared clock instead of being split into per-frame samples.
• Reading the bags: the release splits each sequence into ROS 1 bags by sensor group and by size; every bag of the sequence was read together, in time order, without a ROS install.
• Video encoding: the cameras are re-encoded to Annex-B H.264 without B-frames, one access unit per frame, on each frame's own timestamp.
• Depth: the release records the ZED depth as 32-bit floats in metres, which are carried as 16-bit PNG in millimetres, rounded, with 0 where the ZED reports none.
• Calibration: the calibration is the release's: Kalibr's intrinsics for the five cameras, and its extrinsics chained through the cameras, the inertial unit, the LiDAR, the GPS antenna and the robot's base. Kalibr's camera chain and its camera-to-IMU result map points from one frame into the next, so /tf carries their inverse, each frame's pose in its parent; the LiDAR results are carried as the release gives them, the child's pose in the parent.
• Ground truth: the ground-truth file gives positions only, so it is carried as a position track.
• Left out: the ZED's confidence map and inertial unit, both magnetometers and the GPS receiver's state are not carried.

Dataset Creation

Curation Rationale

[More Information Needed]

Source Data

Data Collection and Processing

Recorded on 2023-07-18 in field 14B of the University of California Riverside's Agricultural Experimental Station, with a Clearpath Jackal carrying the cameras, LiDAR, inertial unit and GPS-RTK receiver described above. The FiftyOne conversion reads the ROS 1 bags without a ROS install and writes one MCAP episode for the sequence; see Parsing decisions above for the changes made.

Who are the source data producers?

The ARCS Lab at the University of California Riverside.

Annotations

Annotation process

The dataset has no FiftyOne label fields. The ground-truth trajectory in the episode is the release's own; how it was produced is [More Information Needed].

Who are the annotators?

[More Information Needed]

Personal and Sensitive Information

[More Information Needed]

Citation

The source release is cited as:

BibTeX:

@inproceedings{teng2023multimodal,
  title={Multimodal Dataset for Localization, Mapping and Crop Monitoring in Citrus Tree Farms},
  author={Teng, Hanzhe and Wang, Yipeng and Song, Xiaoao and Karydis, Konstantinos},
  booktitle={International Symposium on Visual Computing},
  pages={571--582},
  year={2023}
}

APA:

Teng, H., Wang, Y., Song, X., & Karydis, K. (2023). Multimodal dataset for localization, mapping and crop monitoring in citrus tree farms. In International Symposium on Visual Computing (pp. 571-582).

More Information

CitrusFarm is distributed under the Creative Commons Attribution-ShareAlike 4.0 International license (CC-BY-SA-4.0), and this conversion is distributed under the same license.

Changes from the source: one of the release's seven sequences, converted from ROS 1 bags to the FiftyOne MCAP flavor, H.264 encoding of the five camera streams, the ZED depth carried as 16-bit millimetres, the release's calibration carried as camera intrinsics and transforms with Kalibr's inverted, and the confidence map, the ZED inertial unit, the magnetometers and the receiver state left out.

Dataset Card Authors

[More Information Needed]

Dataset Card Contact

[More Information Needed]