Stochastic Segmentation Trees for Multiple Ground Truths

Jake Snell, Richard S. Zemel
Proceedings of the 33rd Conference on Uncertainty in Artificial Intelligence, PMLR R15:750-759, 2017.

Abstract

A strong machine learning system should aim to produce the full range of valid interpretations rather than a single mode in tasks involving inherent ambiguity. This is particularly true for image segmentation, in which there are many sensible ways to partition an image into regions. We formulate a tree-structured probabilistic model, the stochastic segmentation tree, that represents a distribution over segmentations of a given image. We train this model by optimizing a novel objective that quantifies the degree of match between statistics of the model and ground truth segmentations. Our method allows learning of both the parameters in the tree and the structure itself. We demonstrate on two datasets, including the challenging Berkeley Segmentation Dataset, that our model is able to successfully capture the range of ground truths and to produce novel plausible segmentations beyond those found in the data.

Cite this Paper

BibTeX
@InProceedings{pmlr-vR15-snell17a, title = {Stochastic Segmentation Trees for Multiple Ground Truths}, author = {Snell, Jake and Zemel, Richard S.}, booktitle = {Proceedings of the 33rd Conference on Uncertainty in Artificial Intelligence}, pages = {750--759}, year = {2017}, editor = {Elidan, Gal and Kersting, Kristian}, volume = {R15}, series = {Proceedings of Machine Learning Research}, month = {11--15 Aug}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/r15/main/assets/snell17a/snell17a.pdf}, url = {https://proceedings.mlr.press/r15/snell17a.html}, abstract = {A strong machine learning system should aim to produce the full range of valid interpretations rather than a single mode in tasks involving inherent ambiguity. This is particularly true for image segmentation, in which there are many sensible ways to partition an image into regions. We formulate a tree-structured probabilistic model, the stochastic segmentation tree, that represents a distribution over segmentations of a given image. We train this model by optimizing a novel objective that quantifies the degree of match between statistics of the model and ground truth segmentations. Our method allows learning of both the parameters in the tree and the structure itself. We demonstrate on two datasets, including the challenging Berkeley Segmentation Dataset, that our model is able to successfully capture the range of ground truths and to produce novel plausible segmentations beyond those found in the data.}, note = {Reissued by PMLR on 04 October 2026.} }
Endnote
%0 Conference Paper %T Stochastic Segmentation Trees for Multiple Ground Truths %A Jake Snell %A Richard S. Zemel %B Proceedings of the 33rd Conference on Uncertainty in Artificial Intelligence %C Proceedings of Machine Learning Research %D 2017 %E Gal Elidan %E Kristian Kersting %F pmlr-vR15-snell17a %I PMLR %P 750--759 %U https://proceedings.mlr.press/r15/snell17a.html %V R15 %X A strong machine learning system should aim to produce the full range of valid interpretations rather than a single mode in tasks involving inherent ambiguity. This is particularly true for image segmentation, in which there are many sensible ways to partition an image into regions. We formulate a tree-structured probabilistic model, the stochastic segmentation tree, that represents a distribution over segmentations of a given image. We train this model by optimizing a novel objective that quantifies the degree of match between statistics of the model and ground truth segmentations. Our method allows learning of both the parameters in the tree and the structure itself. We demonstrate on two datasets, including the challenging Berkeley Segmentation Dataset, that our model is able to successfully capture the range of ground truths and to produce novel plausible segmentations beyond those found in the data. %Z Reissued by PMLR on 04 October 2026.
APA
Snell, J. & Zemel, R.S.. (2017). Stochastic Segmentation Trees for Multiple Ground Truths. Proceedings of the 33rd Conference on Uncertainty in Artificial Intelligence, in Proceedings of Machine Learning Research R15:750-759 Available from https://proceedings.mlr.press/r15/snell17a.html. Reissued by PMLR on 04 October 2026.

Related Material