Capturing Shape Information with Multi-scale Topological Loss Terms for 3D Reconstruction

Waibel, Dominik J. E.; Atwell, Scott; Meier, Matthias; Marr, Carsten; Rieck, Bastian

doi:10.1007/978-3-031-16440-8_15

Cited by 8 publications

(11 citation statements)

References 16 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…We test the efficacy of our topological loss function by adding it to the SHAPR model and testing it on two biomedical datasets which have been used in the prior work [19,20].…”

Section: Resultsmentioning

confidence: 99%

“…Waibel et al [19] extend the SHAPR model by training the model on a combined loss function of both the DICE loss as well as a regularization term defined by the Wasserstein distance between the persistence diagramstopological descriptors-of the predicted shape and the ground truth. This model outperforms the SHAPR model and provides much better reconstructions than the vanilla SHAPR model.…”

Section: Prior Workmentioning

confidence: 99%

“…where λ is the weight parameter for the topological loss term. This is a similar setup as described by Waibel et al [19], however, our method differs in the details of the topologybased loss L T L (I,Y ) and as a result the efficacy of it as well. Next, we discuss our algorithm to train a neural network by computing the topological loss terms based on ECT, in detail.…”

Section: Overviewmentioning

confidence: 99%

“…Specifically, we design a novel regularization term based on the Euler Characteristic Transform [18], that is computationally efficient, can work with any image size and can be plugged into any neural network. An overview of how our loss function can be used can be seen in Fig 2 . We demonstrate the efficacy of the proposed loss function by plugging it into the SHAPR model and testing it on two bio-medical datasets used in the prior work [19,20]. In the current paper, our key contributions are as follows:…”

Section: Introductionmentioning

confidence: 99%

See 3 more Smart Citations

Euler Characteristic Transform Based Topological Loss for Reconstructing 3D Images from Single 2D Slices

Nadimpalli¹,

Chattopadhyay²,

Rieck³

2023

Preprint

View full text Add to dashboard Cite

The computer vision task of reconstructing 3D images, i.e., shapes, from their single 2D image slices is extremely challenging, more so in the regime of limited data. Deep learning models typically optimize geometric loss functions, which may lead to poor reconstructions as they ignore the structural properties of the shape. To tackle this, we propose a novel topological loss function based on the Euler Characteristic Transform. This loss can be used as an inductive bias to aid the optimization of any neural network toward better reconstructions in the regime of limited data. We show the effectiveness of the proposed loss function by incorporating it into SHAPR, a state-of-the-art shape reconstruction model, and test it on two benchmark datasets, viz., Red Blood Cells and Nuclei datasets. We also show a favourable property, namely injectivity and discuss the stability of the topological loss function based on the Euler Characteristic Transform.

show abstract

“…We test the efficacy of our topological loss function by adding it to the SHAPR model and testing it on two biomedical datasets which have been used in the prior work [19,20].…”

Section: Resultsmentioning

confidence: 99%

Section: Prior Workmentioning

confidence: 99%

Section: Overviewmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Euler Characteristic Transform Based Topological Loss for Reconstructing 3D Images from Single 2D Slices

Nadimpalli¹,

Chattopadhyay²,

Rieck³

2023

Preprint

View full text Add to dashboard Cite

show abstract

“…The binary cross entropy and dice loss used during training are not able to regularize contextual information, but only geometrical information. We explore a possible solution to this by incorporating a topology based loss function ( Waibel et al, 2022 ).…”

Section: Discussionmentioning

confidence: 99%