Code for our paper in CVPR 2021: Indoor Panorama Planar 3D Reconstruction via Divide and Conquer (paper, video)
Download our pretrained models from google drive or dropbox.
- Please resize your images into
512 x 1024. - Follow the preprocessing step here to ensure Mahattan alignment of your 360 images.
- Run our inference script. Examples:
python inference.py --pth ckpt/mp3d.pth --glob static/demo.png --outdir static/mp3d_model_results
To run on a batch of images, you can use --glob "AWESOME_360_IMAGES_DIR/*png"
Here is the visulization example on a held-out data:
python vis_planes.py --img static/demo.png --h_planes static/mp3d_model_results/demo.h_planes.exr --v_planes static/mp3d_model_results/demo.v_planes.exr --mesh
To always visualize all the planes, add --mesh_show_back_face.
@inproceedings{SunHWSC21,
author = {Cheng Sun and
Chi{-}Wei Hsiao and
Ning{-}Hsu Wang and
Min Sun and
Hwann{-}Tzong Chen},
title = {Indoor Panorama Planar 3D Reconstruction via Divide and Conquer},
booktitle = {CVPR},
year = {2021},
}