💡🧠🤔 3D Reconstruction from UAV Images for Plant Phenotyping 🌳🌲🌴

📝 Description:

This repository contains code for a incremental SfM, to reconstruct a point cloud from a field scene.



Input scene	Pointcloud (PIX4Dmapper)	Plant bounding boxes

📒 Journal:

The project is documented in the form of a journal entry. The full document can be found here.

💾 Data:

Please follow instructions to download data from this link.

🔧 Installation:

First, make sure you have Conda installed. Create a virtual environment and install all dependencies from requirements.txt.

If you are keen and have a proper machine, compile COLMAP to be supported by CUDA from the following link

📈 Data Preparation:

If the above provided dataset is used, the images have to be reprocessed to use. Images must be in .png or .jpg format with the max resolution of 2472x1648. Use the following script to resize and convert the images.

First install

pip install rawpy Pillow

Then execute

import os
import rawpy
from PIL import Image

def downscale_CR2(input_path, output_path, target_width=2472, target_height=1648):
    try:
        # Create output directory if it doesn't exist
        os.makedirs(output_path, exist_ok=True)

        # Iterate over each CR2 file in the input directory
        for filename in os.listdir(input_path):
            if filename.endswith(".CR2"):
                input_file = os.path.join(input_path, filename)
                output_file = os.path.join(output_path, filename.replace('.CR2', '_resized.jpg'))

                # Read CR2 file
                with rawpy.imread(input_file) as raw:
                    # Extract the raw image data and convert it to an RGB image
                    rgb = raw.postprocess()

                # Resize the image
                img = Image.fromarray(rgb)
                img = img.resize((target_width, target_height), Image.LANCZOS)

                # Save the resized image as JPEG
                img.save(output_file, format='JPEG')
                print(f"Downscaled and saved: {output_file}")

        print("Downscaling complete.")
    except Exception as e:
        print("Error:", e)


input_path = "input\path"
output_path = "output\path"
downscale_CR2(input_path, output_path)