Tree trunk segmentation with smartphones and neural networks

This repository contains the code for my MRes project as part of the MRes+PhD in AI for Environmental risk at Cambridge University.

In this project we employ a machine learning approach to improve the accuracy of automatic diameter estimates of forest trees using a mobile phone app that runs on a smartphone with a regular camera. We do this by collecting and curating a dataset that allows us to make use of the advantages of depth maps obtained from a LiDAR-equipped phone to refine depth maps obtained from a regular camera using Google ARCore's depth-from-motion algorithm. We train a UNet model with a ResNet34 encoder for a binary image segmentation task of segmenting tree trunks from input depth data, and use the predicted segments to select depths to feed to an existing diameter estimation algorithm.

Data

The data collected for this project is freely available and can be found here. To reproduce the results obtained, download the data and place samples in data/train/samples, data/train/segments, data/test/samples and data/test/segments. Files with a .tiff extension are the target segments, and files without an extension are the inputs. There are also RGB images in the dataset with a .jpeg extension, but these are not used in training and testing.

Experiments

Once you've downloaded the data you can run any of the experiments set up in the configs, for example:

$ python train.py --cfg=config/resnet34_experiment.yaml --o=my/output/dir

Requirements

• Python 3.8+

Getting started

Code implemented in Python 3.8.0

Setting up environment

Create and activate environment

conda env create -f requirements/environment.yml -n env_name 
conda activate env_name 
(env_name)

Project Organization

├── LICENSE
├── Makefile           <- Makefile with commands like `make init` or `make lint-requirements`
├── README.md          <- The top-level README for developers using this project.
|
├── notebooks          <- Jupyter notebooks. Naming convention is a number (for ordering),
|   |                     the creator's initials, and a short `-` delimited description, e.g.
|   |                     `1.0_jqp_initial-data-exploration`.
│   ├── exploratory    <- Notebooks for initial exploration.
│   └── reports        <- Polished notebooks for presentations or intermediate results.
│
├── report             <- Generated analysis as HTML, PDF, LaTeX, etc.
│   ├── figures        <- Generated graphics and figures to be used in reporting
│   └── sections       <- LaTeX sections. The report folder can be linked to your overleaf
|                         report with github submodules.
│
├── requirements       <- Directory containing the requirement files.
│
├── setup.py           <- makes project pip installable (pip install -e .) so src can be imported
├── src                <- Source code for use in this project.
│   ├── __init__.py    <- Makes src a Python module
│   │
│   ├── datasets       <- Scripts defining dataset classes
│   │
│   ├── datamodules    <- Scripts to set up and load data with PyTorch lightning datamodules
│   │
|   |
│   ├── models         <- Scripts to train models and then use trained models to make
│   │                     predictions
│   │
│   └── evaluation     <- Scripts defining functions used in evaluation
│
└── setup.cfg          <- setup configuration file for linting rules

---

Project template created by the Cambridge AI4ER Cookiecutter.