Create INSTALL.md with setup and usage instructions

Added installation instructions and guidelines for reproducibility, CI, and data availability.

Author: angomezu

assets/docs/INSTALL.md

@@ -0,0 +1,106 @@
+## Reproducibility & Installation
+
+### Continuous Integration and Code Quality
+
+This repository enforces consistent coding standards and documentation to support long-term reproducibility and collaborative research.
+
+All Python code is automatically checked using:
+
+- **Ruff** for PEP8 and PEP257 compliance
+- **Pre-commit hooks** to prevent non-compliant code from being committed locally
+- **GitHub Actions CI** to validate code quality on every push and pull request
+
+The CI pipeline runs the following checks:
+
+```bash
+ruff check .
+ruff format --check .
+```
+Pull requests to the main branch are blocked unless all checks pass, ensuring that the repository remains clean, readable, and reproducible over time.
+
+### Data Availability
+
+Due to data access restrictions associated with Oak Ridge National Laboratory (ORNL), the original datasets used in this study are **not publicly available**. Full reproduction of the reported experimental results therefore requires **authorized access** to the Advanced Plant Phenotyping Laboratory (APPL) data.
+
+That said, the codebase is **dataset-agnostic by design**. Any 3D LiDAR point cloud dataset can be used **provided that**:
+- Point clouds are available in **XYZ format** (e.g., `.txt`, `.pcd`, `.ply`)
+- Point-wise semantic labels are provided (or generated) following a compatible annotation scheme
+- The data can be adapted to the expected input format used by the dataset loader
+
+This enables reuse of the pipeline for **methodological experimentation**, architectural benchmarking, and extension to alternative 3D segmentation tasks.
+
+
+### Installation
+
+The main dependencies of the project are listed below.
+
+**Core Requirements**
+- Python ≥ 3.8
+- CUDA ≥ 11.x (optional, but recommended for training)
+- PyTorch + PyTorch Geometric
+- Open3D
+
+
+### Step 1: Clone the Repository and Create Environment
+
+```bash
+git clone https://github.com/angomezu/geometric-deep-learning-plant-organ-segmentation.git
+cd geometric-deep-learning-plant-organ-segmentation
+
+conda create -n plantseg python=3.9 pip
+conda activate plantseg
+```
+
+### Step 2: Install PyTorch
+
+Install PyTorch with CUDA support (adjust CUDA version if needed):
+
+```bash
+pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu118
+```
+
+For CPU-only usage:
+
+```bash
+pip install torch torchvision torchaudio
+```
+
+### Step 3: Install PyTorch Geometric
+
+Install PyTorch Geometric and its dependencies:
+
+```bash
+pip install torch-geometric
+```
+
+If you encounter issues, refer to the official installation guide:
+https://pytorch-geometric.readthedocs.io/en/latest/install/installation.html
+
+### Step 4: Install 3D Processing and ML Dependencies
+
+```bash
+pip install open3d numpy scikit-learn tqdm
+```
+
+
+## Notes on Usage
+
+- Training scripts assume point-wise labeled data
+- Data loaders and feature computation logic are implemented in src/dataset.py
+- Visualization utilities require a functioning OpenGL context (for on-screen rendering)
+
+### Users intending to apply the pipeline to new datasets may need to:
+
+- Adapt the annotation format
+- Update normalization statistics
+- Adjust neighborhood radius and voxelization parameters
+
+---
+
+### Model checkpoints (.pth)
+
+This project uses PyTorch checkpoint files (`.pth`) to store trained model weights.
+Running `python train.py` will save a checkpoint to `models/` (see the filename in `train.py`).
+Update `MODEL_PATH` (evaluation) and `CHECKPOINT` (visualization) to point to your `.pth`.
+
+---