FP4S-Floor-plan-image-segmentation-via-scribble-based-semi-weakly-supervised-learning

FP4S: Floor plan image segmentation via scribble-based semi-weakly-supervised learning

Official code and instructions for "Floor Plan Image Segmentation Via Scribble-Based Semi-Weakly Supervised Learning: A Style and Category-Agnostic Approach"

Introduction

This project introduces a scribble-based semi-weakly-supervised framework for floor plan image segmentation, merging weakly annotated and unlabeled images to boost model robustness and generalizability. This framework benefits from a simplified annotation process while retaining detailed information. Accordingly, we provide a new benchmark dataset for floor plan image parsing covering a wide range of architectural styles and categories.

Dependencies

• Linux
• Pytorch 1.13.1+cu117
• Other required packages are summerized in requirements.txt
• CUDA-supported GPU with at least 24 GB memory size is required for training, and at least 6 GB memory size is required for inference.

Quick start

Download the repository

git clone https://github.com/JanineCHEN/FP4S.git
cd FP4S

Setup virtual environment

Using virtualenvwrapper as an example:

mkvirtualenv FP4S
workon FP4S
pip install torch==1.13.1+cu117 torchvision==0.14.1+cu117 --extra-index-url https://download.pytorch.org/whl/cu117
pip install -r requirements.txt

Dataset

For downloading the dataset, please refer to <a href="https://github.com/JanineCHEN/FP4S/tree/main/dataset">dataset</a>.

Training

For training the FP4S model, please run:

python main.py

For customized configuration, please refer to tools/config.py: If you want to use cyclic learning rate schedular for training, please run:

python main.py --lr_schedule True

If you want to normalize the floor plan images for training, please run:

python main.py --ifNorm True

If you want to leverage pre-trained model weights of the chosen backbone for training, please run:

python main.py --ifpretrain True

For using either Focal loss or abCE loss for training, please run:

python main.py --useFocalLoss True

or

python main.py--useabCELoss True

If you want to use cutmix augmentation, please run:

python main.py --cutmix True

If you want to change the backbone, please run:

python main.py --backbone <the chosen backbone name>

If you want to change the batchsize, please run (notice that if cutmix is set to True, batchsize will be set to 2 as default):

python main.py --batchsize <your preferred batchsize>

Download the checkpoint

For downloading the checkpoints, please refer to <a href="https://github.com/JanineCHEN/FP4S/tree/main/ckpt">ckpt</a>.

Acknowledgement

The computational work for this research was performed on resources of the National Supercomputing Centre, Singapore (https://www.nscc.sg). The data sources used in this study are also gratefully acknowledged. This research was supported by the President's Graduate Fellowship of the National University of Singapore.