PolarDepth: Monocular Transparent Object Depth from Polar-Physics Priors

Wen Dong, Haiyang Mei, Yinglian Ji, Zijun Zhang, Wenyuan Zhang, Pengwei Luo, Bo Dong, Shengfeng He, Xin Yang
Proceedings of the 43rd International Conference on Machine Learning, PMLR 306:25964-25979, 2026.

Abstract

Depth estimation for transparent objects remains a fundamental challenge, as RGB-based cues often fail in regions affected by refraction and light transmission. Polarization provides physically grounded information related to surface orientation and material properties, offering reliable geometric cues even in the absence of texture. In this work, we introduce PolarDepth, a monocular framework that incorporates both RGB and polarization inputs, including the degree and angle of linear polarization (DoLP and AoLP), to estimate dense depth and localize transparent regions. PolarDepth injects polarization-derived physical priors by estimating the refractive index, zenith angle, and azimuth angle from polarization measurements and embedding them into an implicit geometric representation that constrains depth inference in ambiguous transparent regions. To support model development and evaluation, we introduce PTOD, a dataset with synchronized RGB, polarization, and depth data and manually annotated transparent region masks. Experimental results demonstrate that PolarDepth achieves state-of-the-art performance in transparent object depth estimation. The findings highlight the effectiveness of embedding polarization-derived physical priors into learned representations for robust perception in complex visual environments.

Cite this Paper

BibTeX
@InProceedings{pmlr-v306-dong26i, title = {{P}olar{D}epth: Monocular Transparent Object Depth from Polar-Physics Priors}, author = {Dong, Wen and Mei, Haiyang and Ji, Yinglian and Zhang, Zijun and Zhang, Wenyuan and Luo, Pengwei and Dong, Bo and He, Shengfeng and Yang, Xin}, booktitle = {Proceedings of the 43rd International Conference on Machine Learning}, pages = {25964--25979}, year = {2026}, editor = {Zhang, Tong and Dudik, Miroslav and Jaggi, Martin and Agarwal, Alekh and Li, Sharon and Schuurmans, Dale and Zhu, Jerry and Berkenkamp, Felix and Dong, Hanze and Bietti, Alberto}, volume = {306}, series = {Proceedings of Machine Learning Research}, month = {06--11 Jul}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v306/main/assets/dong26i/dong26i.pdf}, url = {https://proceedings.mlr.press/v306/dong26i.html}, abstract = {Depth estimation for transparent objects remains a fundamental challenge, as RGB-based cues often fail in regions affected by refraction and light transmission. Polarization provides physically grounded information related to surface orientation and material properties, offering reliable geometric cues even in the absence of texture. In this work, we introduce PolarDepth, a monocular framework that incorporates both RGB and polarization inputs, including the degree and angle of linear polarization (DoLP and AoLP), to estimate dense depth and localize transparent regions. PolarDepth injects polarization-derived physical priors by estimating the refractive index, zenith angle, and azimuth angle from polarization measurements and embedding them into an implicit geometric representation that constrains depth inference in ambiguous transparent regions. To support model development and evaluation, we introduce PTOD, a dataset with synchronized RGB, polarization, and depth data and manually annotated transparent region masks. Experimental results demonstrate that PolarDepth achieves state-of-the-art performance in transparent object depth estimation. The findings highlight the effectiveness of embedding polarization-derived physical priors into learned representations for robust perception in complex visual environments.} }
Endnote
%0 Conference Paper %T PolarDepth: Monocular Transparent Object Depth from Polar-Physics Priors %A Wen Dong %A Haiyang Mei %A Yinglian Ji %A Zijun Zhang %A Wenyuan Zhang %A Pengwei Luo %A Bo Dong %A Shengfeng He %A Xin Yang %B Proceedings of the 43rd International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2026 %E Tong Zhang %E Miroslav Dudik %E Martin Jaggi %E Alekh Agarwal %E Sharon Li %E Dale Schuurmans %E Jerry Zhu %E Felix Berkenkamp %E Hanze Dong %E Alberto Bietti %F pmlr-v306-dong26i %I PMLR %P 25964--25979 %U https://proceedings.mlr.press/v306/dong26i.html %V 306 %X Depth estimation for transparent objects remains a fundamental challenge, as RGB-based cues often fail in regions affected by refraction and light transmission. Polarization provides physically grounded information related to surface orientation and material properties, offering reliable geometric cues even in the absence of texture. In this work, we introduce PolarDepth, a monocular framework that incorporates both RGB and polarization inputs, including the degree and angle of linear polarization (DoLP and AoLP), to estimate dense depth and localize transparent regions. PolarDepth injects polarization-derived physical priors by estimating the refractive index, zenith angle, and azimuth angle from polarization measurements and embedding them into an implicit geometric representation that constrains depth inference in ambiguous transparent regions. To support model development and evaluation, we introduce PTOD, a dataset with synchronized RGB, polarization, and depth data and manually annotated transparent region masks. Experimental results demonstrate that PolarDepth achieves state-of-the-art performance in transparent object depth estimation. The findings highlight the effectiveness of embedding polarization-derived physical priors into learned representations for robust perception in complex visual environments.
APA
Dong, W., Mei, H., Ji, Y., Zhang, Z., Zhang, W., Luo, P., Dong, B., He, S. & Yang, X.. (2026). PolarDepth: Monocular Transparent Object Depth from Polar-Physics Priors. Proceedings of the 43rd International Conference on Machine Learning, in Proceedings of Machine Learning Research 306:25964-25979 Available from https://proceedings.mlr.press/v306/dong26i.html.

Related Material