MoPR Hackathon Repository

This repository provides a complete pipeline for processing LiDAR point cloud data, performing classification using RandLA-Net, generating Digital Terrain Models (DTMs), and conducting hydrological analysis for drainage network assessment.

---

Overview

The workflow includes:

1. Model Setup (RandLA-Net)
2. Pre-processing of Point Cloud Data
3. Point Cloud Classification
4. Post-processing and Export
5. DTM Generation (COG format)
6. Hydrological Modelling & Drainage Analysis

---

Installation & Build

The model setup and environment configuration should follow the official RandLA-Net PyTorch implementation:

https://github.com/idsia-robotics/RandLA-Net-pytorch

Please ensure all dependencies (PyTorch, CUDA, etc.) are properly installed.

---

Download Pre-trained Model (Required)

Before running classification, download the pre-trained model:

https://iitk-my.sharepoint.com/:u:/g/personal/moonisali20_iitk_ac_in/IQBU27BkqBoxTZGtqK8AJ4yaAbg6nsPaUBPqKKwCjHmElNM?e=R38gZn

Steps:

1. Download the .zip file
2. Extract the contents
3. Place the extracted folder in:

data/saved_models/

---

Dataset Structure (Important)

Each point cloud must be stored in a separate folder following a strict structure:

data/
│
├── pc_id=1/
│   ├── pc.pickle
│   └── metadata/
│       └── metadata.pickle
│
├── pc_id=2/
│   ├── pc.pickle
│   └── metadata/
│       └── metadata.pickle

Requirements

• Each folder must be named as: pc_id=<integer_id>

• Inside each folder:

  • pc.pickle → contains the point cloud data
  • metadata/metadata.pickle → contains metadata information

---

Workflow

1. Pre-processing (LAZ → PKL)

Convert raw .laz point cloud files into .pkl format required by the model:

python pc2pickle.py

---

2. Classification

Run the trained RandLA-Net model for classification:

python test.py

Ensure that:

• Pre-trained model is placed in data/saved_models/
• Dataset follows the required folder structure
• File paths are correctly specified
• Model weights are properly loaded

---

3. Post-processing (PKL → LAZ)

Convert classified .pkl outputs back to .laz format:

python pickle2pc.py

---

Alternative: Direct Processing (Recommended)

You can skip intermediate steps and directly process .laz files:

python test_laz.py

This will:

• Perform classification
• Directly generate classified .laz output

---

4. DTM Generation (COG Format)

Generate Digital Terrain Model (DTM) in Cloud Optimized GeoTIFF (COG) format:

python las2cog.py

---

5. Hydrological Modelling

Perform hydrological analysis and drainage network extraction:

python waterlogging.py

This script performs:

• Depression filling
• Flow direction computation
• Flow accumulation
• Natural drainage network extraction
• Overlay of streams on identified hotspots
• Detection of unconnected hotspots
• Proposal of alternate drainage network
• Save all layers in folder "outputs"

---

Alternative: Hydrological Modelling with Visualisation (Recommended)

You can visualise each layer before saving as output during hydrological analysis and drainage network extraction:

python waterlogging_with_visualisation.py

waterlogging.ipynb is also provided, which you can run in Google Colab or Jupyter Notebook to understand and visualise each layer at different steps of the hydrological analysis.

---

Required Resources

• Computing: GPU-enabled system (recommended)

• Software Stack:

  • PyTorch
  • GDAL
  • WhiteboxTools
  • CloudCompare
  • QGIS

• Data: High-resolution LiDAR point cloud datasets (.laz)

---

Notes

• Ensure coordinate systems are consistent across all processing steps.
• Strictly follow dataset structure to avoid runtime errors.
• Large datasets may require significant memory and processing time.

---