This repository provides code, data and pretrained models for FRI-Net.
[Paper]
- We have updated the model code.
- We have updated the evaluation and the training codes.
Since our code is built upon Roomformer, we follow the environment setup from Roomformer.
# Create conda environment
conda create -n frinet python=3.8
conda activate frinet
# install pytorch and other requried packages
pip install torch==1.9.0+cu111 torchvision==0.10.0+cu111 -f https://download.pytorch.org/whl/torch_stable.html
pip install -r requirements.txt
# Compile the defromable-attention models from deformable-DETR
cd models/ops
sh make.sh
We directly provide the processed data and pretrained checkpoints, please download them and organize them as following:
FRI-Net/
└── data/
└── stru3d/
├── annotations/
├── input/
├── occ/
├── ...
└── checkpoints/
├── pretrained_ckpt.pth
├── pretrained_room_wise_encoder.pth
For details on data preprocessing, please refer to datasets.
Please run the following command to evaluate the model on Structured3D test set:
python eval_stru3d.py --checkpoint ./checkpoints/pretrained_ckpt.pth
You can get the visualized results on the results.
To train FRI-Net on Structured 3D, please run the following command:
python train_stru3d.py --phase=0 --job_name=train_stru3d
python train_stru3d.py --phase=1 --job_name=train_stru3d
python train_stru3d.py --phase=2 --job_name=train_stru3d
If you have any question about this repo, feel free to drop me an email littledaisy20001227@gmail.com
If you find FRI-Net useful in your research, please cite our paper:
@inproceedings{xu2024fri,
title={FRI-Net: Floorplan Reconstruction via Room-wise Implicit Representation},
author={Honghao Xu and Juzhan Xu and Zeyu Huang and Pengfei Xu and Hui Huang and Ruizhen Hu},
booktitle={ECCV}
year={2024}
}
We thank for the following open source projects: