BIPVfinder
BIPVfinder provides deep-learning checkpoints for segmenting building-integrated photovoltaics (BIPV) on building façades from street-level and web imagery. The associated codebase (training + evaluation + processing) lives in the GitHub repository linked below.
- Code: https://github.com/ycdrn/BIPVfinder
- Training dataset repo: https://github.com/ycdrn/bipv_facades
- Checkpoints in this HF repo:
model_ckpts.zip
Model summary
This Hub repository hosts fine-tuned checkpoints used in:
Deep learning for BIPV segmentation on facades: Comparison with human annotations across facade designs (Building and Environment, 2026).
DOI: https://doi.org/10.1016/j.buildenv.2026.114292
Two architectures are referenced in the project:
- SegFormer (semantic segmentation)
- Mask R-CNN (instance segmentation)
Intended use
Primary intended uses:
- Research on façade PV / BIPV recognition and segmentation
- Automated BIPV mask creation as an input to downstream tasks such as PV area estimation and energy-yield estimation (see the accompanying paper/code)
Training data
Models were trained using the BIPV Facades Dataset (curated set of 400 BIPV façade projects; see dataset repository).
- Dataset repo: https://github.com/ycdrn/bipv_facades
How to use
From this repo, download model_ckpts.zip and extract it locally, e.g.:
unzip model_ckpts.zip -d model_ckpts
Then place the downloaded ckpts in the respective folder: /files/03_model_checkpoints/. Please check the paper repository for further details and inference.
Citation
@article{DURAN2026114292,
title = {Deep learning for BIPV segmentation on facades: Comparison with human annotations across facade designs},
journal = {Building and Environment},
pages = {114292},
year = {2026},
issn = {0360-1323},
doi = {10.1016/j.buildenv.2026.114292},
url = {https://www.sciencedirect.com/science/article/pii/S0360132326000983},
author = {Ayca Duran and Pedram Mirabian and Panagiotis Karapiperis and Christoph Waibel and Bernd Bickel and Arno Schlueter}
}