Towards Fewer Annotations: Active Learning via Region Impurity and Prediction Uncertainty for Domain Adaptive Semantic Segmentation

Xie, Binhui; Yuan, Longhui; Li, Shuang; Liu, Chi Harold; Cheng, Xinjing

doi:10.1109/cvpr52688.2022.00790

Cited by 45 publications

(29 citation statements)

References 62 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…And the improvement compared to MADA demonstrates that the new sample selection metric and ST based semi-supervised domain adaptation can effectively address the weaknesses of the previous method. In addition, our method consistently shows better performance than RIPU [73], which demonstrates that selecting a few images to annotate is better than labeling a few regions of each sample, despite the more time latter one costs. The visualization of three example images is displayed in Fig.…”

Section: Imagementioning

confidence: 81%

“…For the UDA task, traditional adversarial-based [16], [39], [42]- [44], prototype-STbased [47]- [49] and the ResNet-101 version of HRDA [51] (which is the current SOTA) UDA methods are included for comparison (Transformer based methods are not listed due to the different feature extraction ability and segmentation performance upperbound). For the active DA task, we compare with the sample-based [21] and region-based [73] approaches using the same amount of annotation. The results of MADA [22] are also listed.…”

Section: Resultsmentioning

confidence: 99%

“…In addition to the sample-wise active learning, pixelwise [72] and region-wise [73] active DA methods have also been proposed for semantic segmentation tasks. However, the annotators have to go through every image for annotating a few image areas, which costs much more time than only labeling a few images.…”

Section: Warmupmentioning

confidence: 99%

See 2 more Smart Citations

Multi-Anchor Active Domain Adaptation for Semantic Segmentation

Ning

Wei

et al. 2021

2021 IEEE/CVF International Conference on Computer Vision (ICCV)

View full text Add to dashboard Cite

Unsupervised domain adaption has been widely adopted in tasks with scarce annotated data. Unfortunately, mapping the target-domain distribution to the source-domain unconditionally may distort the essential structural information of the target-domain data, leading to inferior performance. To address this issue, we firstly propose to introduce active sample selection to assist domain adaptation regarding the semantic segmentation task. By innovatively adopting multiple anchors instead of a single centroid, both source and target domains can be better characterized as multimodal distributions, in which way more complementary and informative samples are selected from the target domain. With only a little workload to manually annotate these active samples, the distortion of the target-domain distribution can be effectively alleviated, achieving a large performance gain. In addition, a powerful semi-supervised domain adaptation strategy is proposed to alleviate the long-tail distribution problem and further improve the segmentation performance. Extensive experiments are conducted on public datasets, and the results demonstrate that the proposed approach outperforms stateof-the-art methods by large margins and achieves similar performance to the fully-supervised upperbound, i.e., 71.4% mIoU on GTA5 and 71.8% mIoU on SYNTHIA. The effectiveness of each component is also verified by thorough ablation studies. Code is available at https://github.com/munanning/MADAv2.

show abstract

Section: Imagementioning

confidence: 81%

Section: Resultsmentioning

confidence: 99%

See 1 more Smart Citation

Multi-Anchor Active Domain Adaptation for Semantic Segmentation

Ning

Wei

et al. 2021

2021 IEEE/CVF International Conference on Computer Vision (ICCV)

View full text Add to dashboard Cite

show abstract

“…However, the selected points ignore the pixel spatial continuity of the image. Recently, Xie et al [70] greatly improved the segmentation performance in the target domain by exploring the consistency of the image space and selecting the most diverse and uncertain image regions. However, most active learning research works ignore the utilization of massive unlabeled data in the target domain, resulting in high labeling costs.…”

Section: Related Workmentioning

confidence: 99%

“…Evaluation metric. As a common practice [45,46,56,70], we report the mean Intersection-over-Union (mIoU) [17] on the Cityscapes validation set. Specifically, we report the mIoU on the shared 19 classes for GTAV → Cityscapes and report the results on 13 (mIoU*) and 16 (mIoU) common classes for SYNTHIA → Cityscapes.…”

Section: Implementation Detailsmentioning

confidence: 99%

Iterative Loop Method Combining Active and Semi-Supervised Learning for Domain Adaptive Semantic Segmentation

Guan¹,

Yang²

2023

Preprint

View full text Add to dashboard Cite

Recently, self-training and active learning have been proposed to alleviate this problem. Self-training can improve model accuracy with massive unlabeled data, but some pseudo labels containing noise would be generated with limited or imbalanced training data. And there will be suboptimal models if human guidance is absent. Active learning can select more effective data to intervene, while the model accuracy can not be improved because the massive unlabeled data are not used. And the probability of querying sub-optimal samples will increase when the domain difference is too large, increasing annotation cost. This paper proposes an iterative loop learning method combining Self-Training and Active Learning (STAL) for domain adaptive semantic segmentation. The method first uses self-training to learn massive unlabeled data to improve model accuracy and provide more accurate selection models for active learning. Secondly, combined with the sample selection strategy of active learning, manual intervention is used to correct the self-training learning. Iterative loop to achieve the best performance with minimal label cost. Extensive experiments show that our method establishes stateof-the-art performance on tasks of GTAV → Cityscapes, SYNTHIA → Cityscapes, improving by 4.9% mIoU and 5.2% mIoU, compared to the previous best method, respectively. Code will be available.

show abstract