Neural Point Processes for Pixel-wise Regression

Chengzhi Shi, Gözde Özcan, Miquel Sirera Perelló, Yuanyuan Li, Nina Iftikhar Shamsi, Stratis Ioannidis
Proceedings of The 28th International Conference on Artificial Intelligence and Statistics, PMLR 258:1567-1575, 2025.

Abstract

We study pixel-wise regression problems with sparsely annotated images. Traditional regression methods based on mean squared error emphasize pixels with labels, leading to distorted predictions in unlabeled areas. To address this limitation, we introduce Neural Point Processes, a novel approach that combines 2D Gaussian Processes with neural networks to leverage spatial correlations between sparse labels on images. This approach offers two key advantages: it imposes smoothness constraints on the model output and enables conditional predictions when sparse labels are available at inference time. Empirical results on synthetic and real-world datasets demonstrate a substantial improvement in mean-squared error and $R^2$ scores, outperforming standard regression techniques. On the real-world dataset COWC, we achieve an $R^2$ of $0.769$ with $81$ out of $40,000$ ($0.2$%) points labeled, while standard regression loss (MSE) results in an $R^2$ of $0.060$.

Cite this Paper

BibTeX
@InProceedings{pmlr-v258-shi25a, title = {Neural Point Processes for Pixel-wise Regression}, author = {Shi, Chengzhi and {\"O}zcan, G{\"o}zde and Perell{\'o}, Miquel Sirera and Li, Yuanyuan and Shamsi, Nina Iftikhar and Ioannidis, Stratis}, booktitle = {Proceedings of The 28th International Conference on Artificial Intelligence and Statistics}, pages = {1567--1575}, year = {2025}, editor = {Li, Yingzhen and Mandt, Stephan and Agrawal, Shipra and Khan, Emtiyaz}, volume = {258}, series = {Proceedings of Machine Learning Research}, month = {03--05 May}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v258/main/assets/shi25a/shi25a.pdf}, url = {https://proceedings.mlr.press/v258/shi25a.html}, abstract = {We study pixel-wise regression problems with sparsely annotated images. Traditional regression methods based on mean squared error emphasize pixels with labels, leading to distorted predictions in unlabeled areas. To address this limitation, we introduce Neural Point Processes, a novel approach that combines 2D Gaussian Processes with neural networks to leverage spatial correlations between sparse labels on images. This approach offers two key advantages: it imposes smoothness constraints on the model output and enables conditional predictions when sparse labels are available at inference time. Empirical results on synthetic and real-world datasets demonstrate a substantial improvement in mean-squared error and $R^2$ scores, outperforming standard regression techniques. On the real-world dataset COWC, we achieve an $R^2$ of $0.769$ with $81$ out of $40,000$ ($0.2$%) points labeled, while standard regression loss (MSE) results in an $R^2$ of $0.060$.} }
Endnote
%0 Conference Paper %T Neural Point Processes for Pixel-wise Regression %A Chengzhi Shi %A Gözde Özcan %A Miquel Sirera Perelló %A Yuanyuan Li %A Nina Iftikhar Shamsi %A Stratis Ioannidis %B Proceedings of The 28th International Conference on Artificial Intelligence and Statistics %C Proceedings of Machine Learning Research %D 2025 %E Yingzhen Li %E Stephan Mandt %E Shipra Agrawal %E Emtiyaz Khan %F pmlr-v258-shi25a %I PMLR %P 1567--1575 %U https://proceedings.mlr.press/v258/shi25a.html %V 258 %X We study pixel-wise regression problems with sparsely annotated images. Traditional regression methods based on mean squared error emphasize pixels with labels, leading to distorted predictions in unlabeled areas. To address this limitation, we introduce Neural Point Processes, a novel approach that combines 2D Gaussian Processes with neural networks to leverage spatial correlations between sparse labels on images. This approach offers two key advantages: it imposes smoothness constraints on the model output and enables conditional predictions when sparse labels are available at inference time. Empirical results on synthetic and real-world datasets demonstrate a substantial improvement in mean-squared error and $R^2$ scores, outperforming standard regression techniques. On the real-world dataset COWC, we achieve an $R^2$ of $0.769$ with $81$ out of $40,000$ ($0.2$%) points labeled, while standard regression loss (MSE) results in an $R^2$ of $0.060$.
APA
Shi, C., Özcan, G., Perelló, M.S., Li, Y., Shamsi, N.I. & Ioannidis, S.. (2025). Neural Point Processes for Pixel-wise Regression. Proceedings of The 28th International Conference on Artificial Intelligence and Statistics, in Proceedings of Machine Learning Research 258:1567-1575 Available from https://proceedings.mlr.press/v258/shi25a.html.

Related Material