RanPaste: Paste Consistency and Pseudo Label for Semisupervised Remote Sensing Image Semantic Segmentation

Wang, Jia-Xin; Chen, Si‐Bao; Ding, Chris; Tang, Jin; Luo, Bin

doi:10.1109/tgrs.2021.3102026

Cited by 35 publications

(18 citation statements)

References 32 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…For the task of road extraction, trained remote sensing images are high resolution and roads in images are usually few. A method randomly pastes part of the labeled image into the unlabeled image was devoloped in [16], which uses RanPaste method as strong perturbation and then generates pseudo labels for weakly enhanced data to train unlabeled samples. To some…”

Section: Related Workmentioning

confidence: 99%

“…Adding perturbation to data is an effective trick to improve the performance and robustness of the model. RanPaste [16] cut out a quarter of labeled image and paste it into another unlabeled image by multiplying the complementary mask, but it gets limited help for road extraction task since the road is slender and sparse in high-resolution remote sensing images. Inspired by RanPaste [16], we propose a novel image mixing method named Foreground Pasting (FP) for road extraction.…”

Section: B Foreground Pasting (Fp)mentioning

confidence: 99%

“…RanPaste [16] cut out a quarter of labeled image and paste it into another unlabeled image by multiplying the complementary mask, but it gets limited help for road extraction task since the road is slender and sparse in high-resolution remote sensing images. Inspired by RanPaste [16], we propose a novel image mixing method named Foreground Pasting (FP) for road extraction. In addition, to randomly cut out the regular square background of a labeled image to paste, all foreground pixels(or road pixels) are also pasted onto another labeled or unlabeled image, so this FP is effective for both fully supervised and semisupervised road extraction training methods.…”

Section: B Foreground Pasting (Fp)mentioning

confidence: 99%

“…Table I shows the experimental segmentation accuracy results comparing our proposed method FMWDCT with other methods on DeepGlobe Dataset when different model structures were adopted. Different row modules stand for different segmentation networks are used, and each line in the row module indicates an experimental figure of this segmentation network combined with some semi-supervised methods, such as CRAUP [40] , RanPaste [16], CPS [10] and FMWDCT.…”

Section: A Road Extraction On Deepglobe Datasetmentioning

confidence: 99%

“…Wang et al proposed a pixel-wise contrastive learning method which facilitates both image-and pixel-level representation learning. Based on RanPaste [16] and CP 2 [17], we take out all the roads of one randomly selected image and paste them into the corresponding position of another image, to increase the proportion of the foreground. And when the loss is calculated in back propagation, we set a hyperparameter to increase the weight of road loss, which can further solve the imbalance problem and improve the performance of network.…”

Section: Introductionmentioning

confidence: 99%

See 4 more Smart Citations

FMWDCT: Foreground Mixup Into Weighted Dual-Network Cross Training for Semisupervised Remote Sensing Road Extraction

You

Wang

Chen

et al. 2022

IEEE J. Sel. Top. Appl. Earth Observations Remote Sensing

Self Cite

View full text Add to dashboard Cite

With the development of deep learning, the application of automatic road extraction has achieved great success. However, the main challenge is how to make full use of a large number of unlabeled images to improve segmentation models and how alleviate sample imbalance in road extraction tasks. In this paper, we propose a novel semi-supervised remote sensing road extraction approach is refined as Foreground Mixup into Weighted Dual-network Cross Training (FMWDCT), which combines labeled images with unlabeled images to extract road from remote sensing images. FMWDCT is composed of Dualnetwork Cross Training (DCT) and Foreground Pasting (FP). DCT is a new semi-supervised training method and FP is an effective data perturbation method for road extraction. We firstly paste the foreground pixels obtained from labeled images into unlabeled images to produce mixed input images. The mixed pseudo labels are then generated by a combination of highconfidence predictions from the augmented network and labeled masks. Finally, the mixed pseudo labels are used to guide another adversarial basic network for cross training, and this basic network is used to smoothly update the augmented network that corresponds to it. The proposed FMWDCT effectively solve the overfitting problem and imbalance problem of positive and negative sample in the case of a few labeled training samples. We demonstrate the effectiveness of our method on three road extraction datasets, and achieve better performance with few labeled data. Extensive experiments show that the proposed semi-supervised method can learn latent information from the unlabeled data to improve performance.

show abstract

Section: Related Workmentioning

confidence: 99%

Section: B Foreground Pasting (Fp)mentioning

confidence: 99%