This repository provides the UNet checkpoint for PVDepth: Panoramic Video Depth Estimation via Geometry-Aware Spatiotemporal Adaptation. It estimates temporally consistent relative inverse depth from equirectangular images and videos.

Usage

bash

GPU_ID=0 bash run_infer_any.sh \
  <input_path> \
  Soon122/PVDepth \
  ./outputs

See the PVDepth repository for installation and usage. The training dataset is available at PanoCARLA.

Citation

bibtex

@inproceedings{song2026pvdepth,
  author    = {Song, Chuanxin and Peng, Peixi},
  title     = {PVDepth: Panoramic Video Depth Estimation via Geometry-Aware Spatiotemporal Adaptation},
  booktitle = {ICML},
  year      = {2026}
}

License

This checkpoint is subject to the upstream licenses of DepthCrafter and Stable Video Diffusion. See the PVDepth repository for details.