GRASP Dataset

GRASP: Generalizable Robotic Adaptation for Sim-to-Real Plant Interaction

Project Website: https://yogesh2850.github.io/grasp

Overview

GRASP is a YOLO-format segmentation dataset designed for real-time corn stalk segmentation in agricultural environments.

In addition to the dataset, GRASP provides pretrained segmentation model weights for multiple YOLO architectures and model scales.

---

Dataset Structure

The dataset is distributed as two compressed archives:

images.tar.gz
labels.tar.gz

After extraction:

images/
├── train/
└── val/

labels/
├── train/
└── val/

Each image has a corresponding annotation file with the same filename.

Example:

images/train/example.jpg
labels/train/example.txt

---

Annotation Format

Annotations follow the YOLO Segmentation format:

<class_id> <x1> <y1> <x2> <y2> ... <xn> <yn>

where:

• class_id is the object category identifier.
• (x, y) pairs define polygon vertices.
• Coordinates are normalized to image dimensions.
• All coordinate values lie in the range [0, 1].

Example:

0 0.460185 0.155208 0.454630 0.155208 0.444444 0.164062 ...

---

Download

Clone the dataset repository:

git lfs install
git clone https://huggingface.co/datasets/yogesh2850/grasp
cd grasp

The dataset is provided as:

images.tar.gz
labels.tar.gz

Extract the archives:

tar -xzf images.tar.gz
tar -xzf labels.tar.gz

After extraction:

images/
├── train/
└── val/

labels/
├── train/
└── val/

---

Pretrained Weights

Pretrained segmentation weights trained on the GRASP dataset are available in:

Trained_weights/

YOLO11 Models

Model	File
YOLO11n	yolov11n.zip
YOLO11s	yolov11s.zip
YOLO11m	yolov11m.zip
YOLO11l	yolov11l.zip

YOLOv8 Models

Model	File
YOLOv8n	yolov8n.zip
YOLOv8s	yolov8s.zip
YOLOv8m	yolov8m.zip
YOLOv8l	yolov8l.zip

YOLO26 Models

Model	File
YOLO26n	yolo26n.zip
YOLO26s	yolo26s.zip
YOLO26m	yolo26m.zip
YOLO26l	yolo26l.zip

Extract the desired archive before inference or fine-tuning.

Example:

unzip yolov11n.zip

Load a pretrained model:

from ultralytics import YOLO

model = YOLO("best.pt")

results = model.predict("image.jpg")

---

Training with Ultralytics YOLO Segmentation

Create a data.yaml file:

path: .

train: images/train
val: images/val

nc: 1

names:
  0: corn_stalk

Train a segmentation model:

pip install ultralytics

yolo segment train \
    model=yolo11n-seg.pt \
    data=data.yaml \
    imgsz=640 \
    epochs=100

---

Python Example

from ultralytics import YOLO

model = YOLO("yolo11n-seg.pt")

model.train(
    data="data.yaml",
    imgsz=640,
    epochs=100
)

---

Project Resources

• Project Website: https://yogesh2850.github.io/grasp
• Dataset Repository: https://huggingface.co/datasets/yogesh2850/grasp

---

License

Please refer to the repository license for usage terms and redistribution conditions.

---

Citation

If you use this dataset or pretrained weights in your research, please cite:

@misc{grasp_dataset,
  title={GRASP: Generalizable Robotic Adaptation for Sim-to-Real Plant Interaction},
  author={Yogesh Chawla},
  year={2026},
  url={https://yogesh2850.github.io/grasp}
}