Implementation of U-Net and SegNet for building segmentation

Last update: Dec 07, 2022

Overview

Specialized project

Created by Katrine Nguyen and Martin Wangen-Eriksen as a part of our specialized project at Norwegian University of Science and Technology (NTNU).

Models

Most of our code and the U-net model is significantly inspired by this project Unet-for-Person-Segmentation. The SegNet model we created on our own based on other implementations of SegNet in Tensorflow.

Data

The model is trained and tested on Massachusetts Buildings Dataset from Kaggle. The original images where 1500X1500 pixels each over an area of 1500x1500 meters (1mx1m resolution). The original 137 images were cropped into 64x64 pixels and images without building were filtered out.

To make the masks compatible with our model the masks was changed from white (255,255,255) labels to greyscale with value 1. This is done in image_fix.py found in the repo.

Folder structure

Images and masks are saved in local directories and used in data.py and test.py. This is of course possible to change, however if you want to use the exact same code you can follow this folder structure.


.
├── ...
├── building-segmentation                # Directory for all images
│   ├── Images                           # Directory for raw images
│   │   ├── cropped_images_train_64      # Directory for cropped images where number specifies resolution, containg .jpg
│   │   ├── cropped_images_train_128     # Directory for cropped images where number specifies resolution, containg .jpg 
│   │   └── ...                          # More directories with other resolutions
│   ├── Masks                            # Directory for all maskes
│   │   ├── cropped_masks_train_64       # Directory for cropped masks where number specifies resolution, containg .jpg
│   │   ├── cropped_masks_train_128      # Directory for cropped masks where number specifies resolution, containg .jpg 
│   │   └── ...                          # More directories with other resolutions
│   └── Test                             # Miscellaneous information
│       ├── test_64                      # Directory for images where number specifies resolution, containing .jpg
│       └── ...                          # More directories with other resolutions
└── ...

# data.py
    images = glob(os.path.join(dataset_path, "images/cropped_images_train_64/*"))
    masks = glob(os.path.join(dataset_path, "masks/cropped_masks_train_64/*"))
    
    # In main:
        dataset_path = "building-segmentation"
    
# test.py
    test_images = glob("building-segmentation/test/test_64/*")

Implementation of U-Net and SegNet for building segmentation

Related tags

Overview

Specialized project

Models

Data

Folder structure

Running the project

Requirements

Training

Testing

Owner

Martin.w-e

Interactive Image Generation via Generative Adversarial Networks

The code for our paper CrossFormer: A Versatile Vision Transformer Based on Cross-scale Attention.

Official code for "Distributed Deep Learning in Open Collaborations" (NeurIPS 2021)

Code for the CVPR2021 paper "Patch-NetVLAD: Multi-Scale Fusion of Locally-Global Descriptors for Place Recognition"

SAGE: Sensitivity-guided Adaptive Learning Rate for Transformers

Source code for paper "Deep Superpixel-based Network for Blind Image Quality Assessment"

Bio-OFC gym implementation and Gym-Fly environment

Geometry-Aware Learning of Maps for Camera Localization (CVPR2018)

A script helps the user to update Linux and Mac systems through the terminal

SpeechBrain is an open-source and all-in-one speech toolkit based on PyTorch.

Self-Supervised Methods for Noise-Removal

meProp: Sparsified Back Propagation for Accelerated Deep Learning

Vowpal Wabbit is a machine learning system which pushes the frontier of machine learning with techniques such as online, hashing, allreduce, reductions, learning2search, active, and interactive learning.

Histocartography is a framework bringing together AI and Digital Pathology

Out-of-Domain Human Mesh Reconstruction via Dynamic Bilevel Online Adaptation

Go from graph data to a secure and interactive visual graph app in 15 minutes. Batteries-included self-hosting of graph data apps with Streamlit, Graphistry, RAPIDS, and more!

TransMVSNet: Global Context-aware Multi-view Stereo Network with Transformers.

Attention mechanism with MNIST dataset

Code for one-stage adaptive set-based HOI detector AS-Net.

Trained on Simulated Data, Tested in the Real World