Decoding the multi-dimensional matrix of agriculture through AI vision.
Tesseract (Hypercube) is a four-dimensional analogue of a cube. Just as a Tesseract extends a 3D cube into a 4th dimension, AgriTesseract extends traditional 2D field phenotyping into higher dimensions:
- Dimension X & Y (Space): Precise subplot segmentation and spatial mapping of every single plant in the field.
- Dimension Z (Depth/Health): Using Custom 3D-CNN to analyze stacked multi-modal layers (RGB + Multispectral + DSM), diving deeper than surface-level imagery.
- Dimension T (Time): Encoding growth stages into grayscale intensity layers, allowing the network to learn temporal sequences and detect growth anomalies.
In Interstellar, the Tesseract allowed access to time as a physical dimension. Similarly, AgriTesseract stacks time and spectrum into a 10-layer hypercube, enabling the detection of subtle growth abnormalities invisible to the naked eye.
- Automated Grid Slicing: Effortlessly slice massive orthomosaic maps into manageable subplots.
- Geo-Referencing: Maintain precise GPS coordinates for every pixel.
- Ridge Detection: Smart algorithms to identify crop rows and furrows automatically.
- Zero-Shot Segmentation: Powered by Segment Anything Model 3 (SAM3), detecting seedlings without extensive retraining.
- Precise Localization: Extract exact centroids and bounding boxes for every individual plant.
- ID Persistence: Assign and track unique IDs for thousands of plants across the field.
- Multi-Modal Stacking: A novel approach stacking RGB, Multispectral, DSM, and Temporal Growth Grayscale into a unified 10-layer tensor.
- Temporal Learning: The modified 3D-CNN learns the "normal" trajectory of plant growth across time sequences encoded in the input layers.
- Anomaly Scoring: Identifies plants with deviant growth patterns (e.g., stunted growth, delayed emergence) by analyzing the reconstruction error of this hyper-dimensional data.
- Modern GUI: Built with PySide6 and QFluentWidgets for a stunning, dark-mode focused experience.
- Data Export: Seamlessly export results to shapefiles, CSVs, or visualize directly on Google Maps satellite layers.
- Real-time Plotting: High-performance rendering of thousands of data points using
pyqtgraph.
(Placeholders for future screenshots - Imagine a sleek, dark-themed interface with glowing green accents)
| Dashboard | Segmentation View |
|---|---|
 Overview of field metrics and project status |
 Real-time SAM3 inference on subplots |
| 3D Analysis | Map Visualization |
|---|---|
 3D-CNN feature space visualization |
 Geo-tagged results on satellite map |
This project is built on the shoulders of giants:
- Core:
Python 3.10+ - GUI Framework:
PySide6(Qt for Python) - UI Components:
QFluentWidgets(Windows 11 Fluent Design style) - Computer Vision:
Ultralytics (SAM3),OpenCV,PyTorch - Data Science:
NumPy,SciPy(Signal Processing),Pandas,GeoPandas - Visualization:
PyQtGraph(High-performance plotting) - Testing:
Pytest
- OS: Linux (Recommended) / Windows 11
- Package Manager:
uv(Fast Python package installer)
-
Clone the repository:
git clone https://github.com/UTokyo-FieldPhenomics-Lab/AgriTesseract.git cd AgriTesseract -
Set up environment with
uv:# Create virtual environment and sync dependencies uv sync -
Run the application:
uv run python launch.py
Contributions are welcome! Please read our Contributing Guide for details on our code of conduct, and the process for submitting pull requests.
This project is licensed under the MIT License - see the LICENSE file for details.
Built with ❤️ and ☕ by UTokyo Field Phenomics Lab
"Decoding nature, one pixel at a time."