gum-tech/horizonnet-resnet50-rnn

HorizonNet ResNet50 + RNN (ST3D)

This repository contains pretrained weights for HorizonNet, a deep learning model for room layout estimation from 360° panorama images.

Model Details

• Architecture: ResNet50 backbone + RNN
• Training Dataset: Structured3D
• Task: Room layout estimation (wall boundaries and corners)
• Input: 360° equirectangular panorama (512x1024)
• Output: Boundary predictions (ceiling/floor) and corner predictions

Usage

import torch
from horizonnet.model import HorizonNet

# Load model
model = HorizonNet(backbone="resnet50", use_rnn=True)
checkpoint = torch.hub.load_state_dict_from_url(
    "https://huggingface.co/gum-tech/horizonnet-resnet50-rnn/resolve/main/resnet50_rnn__st3d.pth",
    map_location="cpu"
)
model.load_state_dict(checkpoint if not isinstance(checkpoint, dict) else checkpoint["state_dict"])
model.eval()

# Run inference
with torch.no_grad():
    y_bon, y_cor = model(panorama_tensor)  # [1, 3, 512, 1024]

Citation

@inproceedings{sun2019horizonnet,
  title={HorizonNet: Learning Room Layout with 1D Representation and Pano Stretch Data Augmentation},
  author={Sun, Cheng and Hsiao, Chi-Wei and Sun, Min and Chen, Hwann-Tzong},
  booktitle={IEEE Conference on Computer Vision and Pattern Recognition (CVPR)},
  year={2019}
}

Original Repository

Original implementation: sunset1995/HorizonNet

License

MIT License - See original repository for details.