DACS: Domain Adaptation via Cross-domain Mixed Sampling

By Wilhelm Tranheden, Viktor Olsson, Juliano Pinto, Lennart Svensson. IEEE Workshop/Winter Conference on Applications of Computer Vision, 2020

458 citations recorded by the source corpus.

Semantic segmentation models based on convolutional neural networks have recently displayed remarkable performance for a multitude of applications. However, these models typically do not generalize well when applied on new domains, especially when going from synthetic to real data. In this paper we address the problem of unsupervised do-main adaptation (UDA), which attempts to train on labelled data from one domain (source domain), and simultaneously learn from unlabelled data in the domain of interest (target domain). Existing methods have seen success by training on pseudo-labels for these unlabelled images. Multiple techniques have been proposed to mitigate low-quality pseudo-labels arising from the domain shift, with varying degrees of success. We propose DACS: Domain Adaptation via Cross-domain mixed Sampling, which mixes images from the two domains along with the corresponding labels and pseudo-labels. These mixed samples are then trained on, in addition to the labelled data itself. We demonstrate the effectiveness of our solution by achieving state-of-the-art results for GTA5 to Cityscapes, a common synthetic-to-real semantic segmentation benchmark for UDA.

Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition, Radiology, Nuclear Medicine and Imaging

View Paper: https://www.semanticscholar.org/paper/277950d4c7f046f2961729ea88bd04fd1b66e7ee

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Computer Vision and Pattern Recognition

In Radiology, Nuclear Medicine and Imaging

Citations in the corpus

DAFormer: Improving Network Architectures and Training Strategies for Domain-Adaptive Semantic Segmentation (2021, methodology): “For a fair comparison, we have re-implemented DACS [15] with our training configuration including learning rate warmup, which achieves 54.”
HRDA: Context-Aware High-Resolution Domain-Adaptive Semantic Segmentation (2022, unknown): “After each training step t , the teacher model g ϕ is updated with the exponentially moving average of the weights of f θ , implementing a temporal ensemble to stabilize pseudo-labels, which is a common strategy in semi-supervised learning [18,57,62] and UDA [1,41,63] Further, DAFormer [29] uses con”
Undoing the Damage of Label Shift for Cross-domain Semantic Segmentation (2022, methodology): “Recently, selfsupervised methods, including self-training [16, 17, 32, 49, 51, 52, 54, 55] and semantic consistency [26, 41, 47], have been shown to significantly improve the cross-domain semantic segmentation performance.”
MIC: Masked Image Consistency for Context-Enhanced Domain Adaptation (2022, methodology): “For semantic segmentation, we follow [34, 35, 85] and utilize the ratio of pixels exceeding a threshold τ of the maximum softmax probability”
BAPA-Net: Boundary Adaptation and Prototype Alignment for Cross-domain Semantic Segmentation (2021, methodology): “To assign labels to the mixed image Xm, [49] employ a Mean Teacher (MT) [48] model to assign pseudo-labels to the target image.”
Tackling Catastrophic Forgetting and Background Shift in Continual Semantic Segmentation (2021, background): “Image mixing is popular for classification [93], [94], [95], [96], [97], [98] yet, to the best of our knowledge, sees limited use for semantic segmentation [99], [100], [101], [102], [103], and has never been considered to design memory-efficient rehearsal learning systems.”