Batch Mode Active Learning for Semantic Segmentation Based on Multi-Clue Sample Selection

Tan, Yao; Yang, Liu; Hu, Qinghua; Du, Zhibin

doi:10.1145/3357384.3357968

Cited by 8 publications

(3 citation statements)

References 27 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…[13] and [20] have verified this method is quite competitive in image classification and segmentation tasks. -QBC (Uncertainty): previous methods designed for semantic segmentation, like [4,11,22,23], all use a group of models to measure uncertainty. We use the efficient MC dropout to represent these methods and report the best performance out of both the max-entropy and variation-ratio acquisition functions.…”

Section: Evaluated Methodsmentioning

confidence: 99%

“…The latter views the AL process as an approximation of the entire data distribution and query samples to increase the data diversity, such as Core-set [14] and VAAL [15], which can be directly used in semantic segmentation. There are also some methods specially designed for semantic segmentation, which can also be divided into two groups: image-level [4,11,19,22] and region-level [23,24,20].…”

Section: Related Workmentioning

confidence: 99%

“…[19] adopt QBC strategy and propose a cost-sensitive active learning method for intracranial hemorrhage detection. [22] build a batch mode multi-clue method, incorporating edge information with QBC strategy and graph-based representativeness. All of them are based on a group of models and time-consuming when querying a large unlabeled data pool.…”

Section: Related Workmentioning

confidence: 99%

See 2 more Smart Citations

DEAL: Difficulty-aware Active Learning for Semantic Segmentation

Xie

Feng

Chen

et al. 2020

Preprint

View full text Add to dashboard Cite

Active learning aims to address the paucity of labeled data by finding the most informative samples. However, when applying to semantic segmentation, existing methods ignore the segmentation difficulty of different semantic areas, which leads to poor performance on those hard semantic areas such as tiny or slender objects. To deal with this problem, we propose a semantic Difficulty-awarE Active Learning (DEAL) network composed of two branches: the common segmentation branch and the semantic difficulty branch. For the latter branch, with the supervision of segmentation error between the segmentation result and GT, a pixel-wise probability attention module is introduced to learn the semantic difficulty scores for different semantic areas. Finally, two acquisition functions are devised to select the most valuable samples with semantic difficulty. Competitive results on semantic segmentation benchmarks demonstrate that DEAL achieves state-of-the-art active learning performance and improves the performance of the hard semantic areas in particular.

show abstract

Section: Evaluated Methodsmentioning

confidence: 99%