MungSeedAI

Companion code for:

Integrative Genomics and Deep Learning-Based Phenotyping Reveal the Genetic Architecture of Seed Traits in Mung bean (Vigna radiata L.) Boddepalli, Jubery, Cannon, Farmer, Dutta, Ganapathysubramaniam, and Singh (2026)

Uses Segment Anything Model (SAM) to segment seeds from flatbed scanner images and extract per-seed morphological and color traits (area, axis lengths, aspect ratio, mean hue). Extracted traits feed downstream GWAS and genomic analyses in the companion manuscript.

---

Repository Structure

MungSeedAI/
├── sam_for_seed.py       # Main script
├── requirements.txt
├── LICENSE
└── examples/
    ├── input/            # Example scanner image
    └── output/           # Example outputs

---

Installation

conda create -n mungseed python=3.10 -y && conda activate mungseed
pip install -r requirements.txt
pip install git+https://github.com/facebookresearch/segment-anything.git

Download the SAM ViT-L checkpoint and place it alongside the script:

wget https://dl.fbaipublicfiles.com/segment_anything/sam_vit_l_0b3195.pth

Other checkpoints (vit_h, vit_b) are available at the SAM model zoo. Update sam_checkpoint and model_type in the script accordingly.

---

Usage

Edit the three variables at the top of sam_for_seed.py:

Variable	Description
input_directory_1	Directory of input .jpg/.png scan images
output_directory	Directory for output files
sam_checkpoint	Path to downloaded SAM .pth file

Then run:

python sam_for_seed.py

Outputs per image:

File	Description
<name>_color_mask.png	SAM masks with random colors
<name>_filtered_mask.png	IQR-filtered masks applied to original image
<name>_seed_properties.csv	Per-seed traits: area, axis lengths, aspect ratio, mean hue

Hardware: GPU strongly recommended (≥ 8 GB VRAM for ViT-L). Falls back to CPU automatically.

---

Example

Input	Color mask	Filtered mask

---

Citation

@article{boddepalli2026mungseedai,
  title  = {Integrative Genomics and Deep Learning-Based Phenotyping Reveal the Genetic Architecture of Seed Traits in Mung bean (\textit{Vigna radiata} L.)},
  author = {Boddepalli, Venkata Naresh and Jubery, Talukder Zaki and Cannon, Steven B. and Farmer, Andrew and Dutta, Somak and Ganapathysubramaniam, Baskar and Singh, Arti},
  year   = {2026}
}

Please also cite SAM: Kirillov et al., Segment Anything, arXiv:2304.02643 (2023).

---

License

MIT — see LICENSE.

Contact

Code: Zaki Jubery (znjubery@iastate.edu) and Baskar Ganapahtysubramanian (baskarg@iastate.edu) · Manuscript: Arti Singh (artisingh@iastate.edu)