Point-Supervised Segmentation Of Microscopy Images And Volumes Via Objectness Regularization

Li, Shijie; Dey, Neel; Bermond, Katharina; Emde, Leon von der; Curcio, Christine A.; Ach, Thomas; Gerig, Guido

doi:10.1109/isbi48211.2021.9433963

Cited by 7 publications

(12 citation statements)

References 18 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Other works distinguish between different levels of weakly supervised annotations, such as bounding boxes [9], scribbles [17], points [15,34], image labels [23], pixel-level pseudo labels generated with class activation maps [12,25,27], and also a text-driven semantic segmentation [18]. While fully-labelled data is limited, point labels are also used in instance segmentation methods, such as [13] introduced a novel learning scheme in instance segmentation with point labels and [14] proposed point-level instance segmentation with two branch network such as localisation and embedding branch.…”

Section: Related Workmentioning

confidence: 99%

“…Bearman et al [3] proposed a methodology by incorporating objectness potential in the training loss function in segmentation models with image and point-level annotations. Li et al [15] utilised an objectness prior similar to [3] but instead of a convolutional neural network (CNN) output they utilize distances in the pixel and colour space, meaning that the further away in the image and the more different the colour, the objectness decreases. Zhang et al [34] proposed a contrast-based variational model [22] for semantic segmentation that supports reliable complementary supervision to train a model for histopathology images.…”

Section: Related Workmentioning

confidence: 99%

“…Their proposed method incorporates the correlation between each location in the image and annotations of in-target and out-of-target classes. The weak supervision part of our research is inspired by [3,15,34], as we have only a single point for each tree, we use point labels in combination with denser background information while considering an objectness prior.…”

Section: Related Workmentioning

confidence: 99%

“…Originally proposed in [3], pseudo-labels are derived from model predictions that are pretrained on another dataset with the same task. We derive pseudo-labels from an adapted watershed algorithm [15] to increase the extent of the object being sensed by the model for point-level supervision without requiring a pretrained model on the same task. Unlike common weak supervision scenarios that assumes sparse but fully annotated data, our incomplete annotations can lead to an incorrect objectness prior.…”

Section: Introductionmentioning

confidence: 99%

See 3 more Smart Citations

Predicting urban tree cover from incomplete point labels and limited background information

Zhang,

Kariryaa,

Guthula

et al. 2023

Proceedings of the 1st ACM SIGSPATIAL International Workshop on Advances in Urban-Ai

View full text Add to dashboard Cite

Trees inside cities are important for the urban microclimate, contributing positively to the physical and mental health of the urban dwellers. Despite their importance, often only limited information about city trees is available. Therefore in this paper, we propose a method for mapping urban trees in high-resolution aerial imagery using limited datasets and deep learning. Deep learning has become best-practice for this task, however, existing approaches rely on large and accurately labelled training datasets, which can be difficult and expensive to obtain. However, often noisy and incomplete data may be available that can be combined and utilized to solve more difficult tasks than those datasets were intended for.This paper studies how to combine accurate point labels of urban trees along streets with crowd-sourced annotations from an open geographic database to delineate city trees in remote sensing images, a task which is challenging even for humans. To that end, we perform semantic segmentation of very high resolution aerial imagery using a fully convolutional neural network.The main challenge is that our segmentation maps are sparsely annotated and incomplete. Small areas around the point labels of the street trees coming from official and crowd-sourced data are marked as foreground class. Crowd-sourced annotations of streets, buildings, etc. define the background class. Since the tree data is incomplete, we introduce a masking to avoid class confusion.Our experiments in Hamburg, Germany, showed that the system is able to produce tree cover maps, not limited to trees along streets, without providing tree delineations. We evaluated the method on manually labelled trees and show that performance drastically deteriorates if the open geographic database is not used.

show abstract

Section: Related Workmentioning

confidence: 99%

Section: Related Workmentioning

confidence: 99%

Section: Related Workmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Predicting urban tree cover from incomplete point labels and limited background information

Zhang,

Kariryaa,

Guthula

et al. 2023

Proceedings of the 1st ACM SIGSPATIAL International Workshop on Advances in Urban-Ai

View full text Add to dashboard Cite

show abstract

“…Sobel filter to obtain pseudo edge maps for regularization in training [3,4], or exploits objectness to generate pseudo mask for supervision [5]. Those methods, however, typically share semantic and instance representations at the pixel level, and thus have difficulty in handling crowded nuclei instances due to the lack of expressive instance-aware representations.…”

Section: Introductionmentioning

confidence: 99%

Weakly Supervised Nuclei Segmentation via Instance Learning

Liu¹,

Qian²,

He³

2022

Preprint

View full text Add to dashboard Cite

Weakly supervised nuclei segmentation is a critical problem for pathological image analysis and greatly benefits the community due to the significant reduction of labeling cost. Adopting point annotations, previous methods mostly rely on less expressive representations for nuclei instances and thus have difficulty in handling crowded nuclei. In this paper, we propose to decouple weakly supervised semantic and instance segmentation in order to enable more effective subtask learning and to promote instance-aware representation learning. To achieve this, we design a modular deep network with two branches: a semantic proposal network and an instance encoding network, which are trained in a two-stage manner with an instance-sensitive loss. Empirical results show that our approach achieves the state-of-the-art performance on two public benchmarks of pathological images from different types of organs. Our code is available at https://github.com/weizhenFrank/WeakNucleiSeg.

show abstract