cornvision

Offline corn kernel detection and row counting from ear images.

Installation

pip install -e .

Model weights are downloaded automatically from Hugging Face on first run and cached at ~/.cornvision/model.pth.

Usage

GUI

cornvision

The GUI supports both single and batch inference:

Queue items are colour-coded: ○ pending, ↻ running, ✓ done, ✗ error.

Python API

from cornvision import CornAnalyzer

analyzer = CornAnalyzer()          # loads model, downloads weights if needed
result = analyzer.analyze("corn.jpg")

print(result.total_kernels)        # int
print(result.total_rows)           # int
print(result.kernels_per_row)      # list[int]
print(result.inference_time_sec)   # float

result.save_viz("annotated.png")   # save annotated image to disk

CornAnalyzer loads the model once and can analyze multiple images:

for path in image_paths:
    result = analyzer.analyze(path)

CornAnalyzer(device="auto")

AnalysisResult fields

Dependencies

tkinter is used for the GUI and ships with the standard Python installer.

Project layout

src/cornvision/
├── __init__.py       # public API: CornAnalyzer, AnalysisResult
├── _gui.py           # tkinter GUI (entry point)
├── _analyzer.py      # CornAnalyzer — model load + analyze()
├── _inference.py     # end-to-end inference pipeline
├── _clustering.py    # kernel row clustering
├── _image_proc.py    # image pre/post-processing helpers
├── _model.py         # model architecture + weight loading
├── _download.py      # weight download + cache path
├── _config.py        # shared constants
└── _result.py        # AnalysisResult dataclass