Learning to Count via Unbalanced Optimal Transport

Ma, Zhiheng; Wei, Xing; Hong, Xiaopeng; Lin, Huei-Yung; Qiu, Yunfeng; Gong, Yi

doi:10.1609/aaai.v35i3.16332

Cited by 60 publications

(7 citation statements)

References 43 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…We use the mathematical tool of optimal transport (OT) to solve for Π . OT finds the most efficient way of moving mass between two distributions [30], and has previously been applied to single-cell alignment [11, 40] and spatial transcriptomics alignment [24, 45]. We seek to transport the mass of slice 𝒮 1 to 𝒮 2 where the mass is represented as a distribution over each slice’s spots.…”

Section: Methodsmentioning

confidence: 99%

DeST-OT: Alignment of Spatiotemporal Transcriptomics Data

Halmos,

Liu,

Gold

et al. 2024

Preprint

View full text Add to dashboard Cite

Spatially resolved transcriptomics (SRT) measures mRNA transcripts at thousands of locations within a tissue slice, revealing spatial variations in gene expression and distribution of cell types. In recent studies, SRT has been applied to tissue slices from multiple timepoints during the development of an organism. Alignment of this spatiotemporal transcriptomics data can provide insights into the gene expression programs governing the growth and differentiation of cells over space and time. We introduce DeST-OT (Developmental SpatioTemporal Optimal Transport), a method to align SRT slices from pairs of developmental timepoints using the framework of optimal transport (OT). DeST-OT uses semi-relaxed optimal transport to precisely model cellular growth, death, and differentiation processes that are not well-modeled by existing alignment methods. We demonstrate the advantage of DeST-OT on simulated slices. We further introduce two metrics to quantify the plausibility of a spatiotemporal alignment: a growth distortion metric which quantifies the discrepancy between the inferred and the true cell type growth rates, and a migration metric which quantifies the distance traveled between ancestor and descendant cells. DeST-OT outperforms existing methods on these metrics in the alignment of spatiotemporal transcriptomics data from the development of axolotl brain.

show abstract

Section: Methodsmentioning

confidence: 99%

DeST-OT: Alignment of Spatiotemporal Transcriptomics Data

Halmos,

Liu,

Gold

et al. 2024

Preprint

View full text Add to dashboard Cite

show abstract

“…propose to use optimal transport to match the distributions of the point annotations and the density maps [38]. In [23], the unbalanced optimal transport is introduced to quantify the discrepancy between predicted density maps and point annotations. S3 further proposes semibalanced Sinkhorn divergence to solve the limitations of amount constraint and entropic bias [12].…”

Section: Related Work 21 Fully-supervised Crowd Countingmentioning

confidence: 99%

Semi-supervised Crowd Counting via Density Agency

Lin

Hong

et al. 2022

Proceedings of the 30th ACM International Conference on Multimedia

Self Cite

View full text Add to dashboard Cite

In this paper, we propose a new agency-guided semi-supervised counting approach. First, we build a learnable auxiliary structure, namely the density agency to bring the recognized foreground regional features close to corresponding density sub-classes (agents) and push away background ones. Second, we propose a densityguided contrastive learning loss to consolidate the backbone feature extractor. Third, we build a regression head by using a transformer structure to refine the foreground features further. Finally, an efficient noise depression loss is provided to minimize the negative influence of annotation noises. Extensive experiments on four challenging crowd counting datasets demonstrate that our method achieves superior performance to the state-of-the-art semisupervised counting methods by a large margin. Code is available. CCS CONCEPTS• Computing methodologies → Computer vision; Semi-supervised learning settings; Learning from implicit feedback.

show abstract

“…BL [3] introduces a novel loss function based on Bayes theorem, allowing the model to learn directly from the ground-truth point map. Out of the same purpose, several other works have incorporated optimal transport [4], [30]- [32] theory into crowd counting. Moreover, to enhance model performance, some approaches have designed to leverage auxiliary tasks such as foreground classification [33]- [35], depth estimation [24] and perspective estimation [25].…”

Section: A Fully-supervised Crowd Countingmentioning

confidence: 99%

Counting animals in aerial images with a crowd counting model

Qian

Humphries²,

Trathan

et al. 2022

Preprint

View full text Add to dashboard Cite

1. Animal abundance estimation is increasingly based on drone or aerial survey photography. Manual post-processing has been used extensively, however volumes of such data are increasing, necessitating some level of automation, either for complete counting, or as a labour-saving tool. Any automated processing can be challenging when using the tools on species that nest in close formation such as Pygoscelid penguins. 2. We present here an adaptation of state-of-the-art crowd-counting methodologies for counting of penguins from aerial photography. 3. The crowd-counting model performed significantly better in terms of model performance and computational efficiency than standard Faster RCNN deep-learning approaches and gave an error rate of only 0.8 percent. 4. Crowd-counting techniques as demonstrated here have the ability to vastly improve our ability to count animals in tight aggregations, which will demonstrably improve monitoring efforts from aerial imagery.

show abstract

Learning to Count via Unbalanced Optimal Transport

Cited by 60 publications

References 43 publications

DeST-OT: Alignment of Spatiotemporal Transcriptomics Data

DeST-OT: Alignment of Spatiotemporal Transcriptomics Data

Semi-supervised Crowd Counting via Density Agency

Counting animals in aerial images with a crowd counting model

Contact Info

Product

Resources

About