Sesame Plant Segmentation Dataset: A YOLO Formatted Annotated Dataset

AI summary

Overview

Research area: Computer vision for agriculture — specifically instance/semantic segmentation datasets for crop plants, applied here to sesame.

Technical level: Intermediate. The paper assumes familiarity with YOLO annotation formats, segmentation masks, and standard object detection metrics (precision, recall, mAP at IoU thresholds), but its core contribution — a dataset — is accessible to a general audience.

Scope: The paper introduces and benchmarks an open-source, YOLO-formatted, pixel-level annotated image dataset of sesame plants photographed on farms in Katsina State, Nigeria.

Note: this summary is based solely on the abstract; the full text was not available, so details such as annotation guidelines, training hyperparameters, hardware, and comparisons to other datasets are not covered.

What This Paper Is About

Building AI models that can identify and outline crops in real farm photographs requires annotated image data, and such data is scarce for many specific crops — sesame among them. This paper addresses that gap by releasing a labeled image dataset of sesame plants at early growth stages, captured on actual Nigerian farms and annotated at the pixel level rather than with simple bounding boxes. The authors then check whether a standard, widely used vision model (YOLOv8) can actually learn useful detection and segmentation from the data.

Key Contributions

1. An open-source sesame plant segmentation dataset split into 206 training images, 43 validation images, and 43 test images, formatted for YOLO-compatible segmentation workflows.
2. A documented field-collection and annotation pipeline: images captured with a high-resolution mobile camera on farms in Jirdede, Daura Local Government Area, Katsina State, Nigeria, under varying environmental conditions and at early plant growth stages.
3. Pixel-level ground truth generated semi-automatically: annotation performed with Segment Anything Model version 2 (SAM 2) but supervised by farmers, pairing model assistance with domain expertise.
4. A baseline benchmark: the authors trained and evaluated Ultralytics YOLOv8 on the dataset for both bounding-box detection and segmentation, reporting the resulting metrics.

Main Findings

• Detection performance: using Ultralytics YOLOv8 for bounding-box detection, the model reached 79% recall, 79% precision, 84% mean average precision at IoU 0.50, and 58% mean average precision averaged from IoU 0.50 to 0.95.
• Segmentation performance: for pixel-level segmentation, the model reached 82% recall, 77% precision, 84% mean average precision at IoU 0.50, and 52% mean average precision from IoU 0.50 to 0.95.
• Detection versus segmentation trade-off: recall was higher for segmentation (82% vs. 79%) while the stricter mAP from 0.50 to 0.95 was higher for detection (58% vs. 52%), indicating that precise mask localization is harder than box localization on this data.
• Dataset composition and conditions: the corpus is small (292 images total) and deliberately varied — early growth stages, different environmental conditions, real farm settings rather than controlled scenes.
• Claimed novelty: the authors position the dataset as a novel contribution to sesame-focused agricultural vision datasets in Nigeria.
• What is not reported in the abstract: there is no comparison to other datasets or competing models, no per-class or per-growth-stage breakdown, no information on the number of annotated plant instances, and no inter-annotator agreement analysis. Any such details would be in the full paper.

Methodology in Plain English

The authors went into the field and photographed sesame plants at an early growth stage using a high-resolution mobile phone camera, working on farms in a specific locality in Katsina State, Nigeria. Rather than drawing boxes around plants by hand, they used an AI segmentation tool (SAM 2) to propose pixel-level outlines, with farmers — the people who actually know the crop — checking and supervising the annotations so the labels reflect real plants rather than the model's guesses. The labeled images were then converted into the YOLO segmentation format and split into training, validation, and test sets. Finally, they trained the standard Ultralytics YOLOv8 model on the training data and measured how well it detected plants as boxes and outlined them as masks, using standard precision, recall, and mean average precision metrics.

The abstract does not describe the training configuration (epochs, image resolution, augmentation), hardware, annotation guidelines, or how the train/validation/test split was constructed.

Why This Matters

Impact on research: Agricultural computer vision is dominated by a handful of well-resourced crops; datasets for crops like sesame — particularly from African smallholder contexts — are rare. Releasing a YOLO-formatted, pixel-annotated dataset lowers the barrier for other researchers to build, compare, and reproduce sesame-specific vision models, and the reported YOLOv8 numbers give a baseline against which future work can be measured.

Real-world applications (as stated in the abstract):

• Plant monitoring — tracking sesame plant presence and extent across a field over time.
• Yield estimation — using plant coverage and counts as an input to predicting how much a field will produce.
• Agricultural research — providing standardized, machine-readable observations of sesame at early growth stages for agronomic study.
• Developing and benchmarking agricultural AI models — the dataset's stated purpose is to support AI development for agricultural applications more broadly.

Industry relevance: Agri-tech companies building precision-farming tools, crop-scouting apps, or field robotics need crop-specific labeled data before any model can be deployed. Because the dataset uses the widely adopted YOLO segmentation format, it plugs directly into existing toolchains, and because it comes from real farms rather than controlled settings, it reflects the messy conditions commercial systems must handle. The farmer-supervised, SAM 2-assisted annotation approach also illustrates a practical, lower-cost labeling workflow that industry teams could adopt for other crops.

Future Directions

• Scaling and diversifying the dataset: with 292 images in total, expanding to more plants, more locations, and additional growth stages would test whether performance holds beyond the original site and season.
• Comparing models and annotation strategies: the abstract reports only a YOLOv8 baseline; benchmarking other detection and segmentation architectures, and evaluating how much the SAM 2 plus farmer supervision workflow contributed to label quality, are natural next steps.
• Moving toward class distinction and decision-making: a single sesame class is detected here — distinguishing sesame from weeds, or from other crops, would make the data more directly useful for weed management or stand assessment.
• Validating field impact: the abstract claims support for plant monitoring and yield estimation, but no study is presented linking model outputs to actual agronomic measurements such as yield; closing that loop is an open question.

Target Audience

Agricultural AI and computer vision researchers looking for crop-specific segmentation data; dataset creators interested in semi-automatic annotation workflows combining a foundation segmentation model with human expert verification; agronomists and agri-tech developers working on sesame or smallholder farming systems in Nigeria and similar regions; and graduate students or practitioners who need a small, ready-to-use YOLO-segmentation dataset to prototype detection and segmentation pipelines.

Authors’ abstract

This paper presents the Sesame Plant Segmentation Dataset, an open source annotated image dataset designed to support the development of artificial intelligence models for agricultural applications, with a specific focus on sesame plants. The dataset comprises 206 training images, 43 validation images, and 43 test images in YOLO compatible segmentation format, capturing sesame plants at early growth stages under varying environmental conditions. Data were collected using a high resolution mobile camera from farms in Jirdede, Daura Local Government Area, Katsina State, Nigeria, and annotated using the Segment Anything Model version 2 with farmer supervision. Unlike conventional bounding box datasets, this dataset employs pixel level segmentation to enable more precise detection and analysis of sesame plants in real world farm settings. Model evaluation using the Ultralytics YOLOv8 framework demonstrated strong performance for both detection and segmentation tasks. For bounding box detection, the model achieved a recall of 79 percent, precision of 79 percent, mean average precision at IoU 0.50 of 84 percent, and mean average precision from 0.50 to 0.95 of 58 percent. For segmentation, it achieved a recall of 82 percent, precision of 77 percent, mean average precision at IoU 0.50 of 84 percent, and mean average precision from 0.50 to 0.95 of 52 percent. The dataset represents a novel contribution to sesame focused agricultural vision datasets in Nigeria and supports applications such as plant monitoring, yield estimation, and agricultural research.