Domain Adaptation for Structured Output via Discriminative Patch Representations

By Yi-Hsuan Tsai, Kihyuk Sohn, Samuel Schulter, Manmohan Chandraker. IEEE International Conference on Computer Vision, 2019

334 citations recorded by the source corpus.

Predicting structured outputs such as semantic segmentation relies on expensive per-pixel annotations to learn supervised models like convolutional neural networks. However, models trained on one data domain may not generalize well to other domains without annotations for model finetuning. To avoid the labor-intensive process of annotation, we develop a domain adaptation method to adapt the source data to the unlabeled target domain. We propose to learn discriminative feature representations of patches in the source domain by discovering multiple modes of patch-wise output distribution through the construction of a clustered space. With such representations as guidance, we use an adversarial learning scheme to push the feature representations of target patches in the clustered space closer to the distributions of source patches. In addition, we show that our framework is complementary to existing domain adaptation techniques and achieves consistent improvements on semantic segmentation. Extensive ablations and results are demonstrated on numerous benchmark datasets with various settings, such as synthetic-to-real and cross-city scenarios.

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

View Paper: https://www.semanticscholar.org/paper/5173ce808d4c79fd66db4410d25c577e32d058dc

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

Progressive Domain Adaptation for Object Detection (2019, background): “[29, 30] adopt adversarial learning in the structured output space for solving domain adaptation on semantic segmentation.”
ACDC: The Adverse Conditions Dataset with Correspondences for Semantic Driving Scene Understanding (2021, background): “A major class of UDA works employs adversarial domain adaptation to implicitly align the source and target domains at the level of pixels and/or features [7, 15, 26, 41, 43, 44, 46, 48, 60].”
HRDA: Context-Aware High-Resolution Domain-Adaptive Semantic Segmentation (2022, neutral): “The target loss can be defined according to the used training strategies such as adversarial training [64, 65, 69] or self-training [29, 48,63,87,88,99].”
DAFormer: Improving Network Architectures and Training Strategies for Domain-Adaptive Semantic Segmentation (2021, methodology): “Adversarial training methods aim to align the distributions of source and target domain at input [21, 30], feature [31, 72], output [72, 76], or patch level [73] in a GAN framework [19, 22].”
Learning from Extrinsic and Intrinsic Supervisions for Domain Generalization (2020, background): “The general idea of domain adaption is to align the source domain and target domain distributions in the input level [3, 19], semantic feature level [9, 34], or output space level [5, 39, 40, 43, 45].”
Unsupervised Intra-Domain Adaptation for Semantic Segmentation Through Self-Supervision (2020, background): “For the particular task of semantic segmentation, adversarial learningbased UDA approaches demonstrate efficiency in aligning features at the image [16, 11] or output [26, 25] level.”
Source-Free Domain Adaptation for Semantic Segmentation (2021, methodology): “To reduce the cross-domain discrepancy, numerous UDA methods [19, 42, 43, 34] focus on distribution consistency by introducing adversarial learning.”
Classes Matter: A Fine-grained Adversarial Approach to Cross-domain Semantic Segmentation (2020, neutral): “AdaptPatch [31] 82.”
Prototypical Pseudo Label Denoising and Target Structure Learning for Domain Adaptive Semantic Segmentation (2021, methodology): “These methods could be divided into two categories: 1) domain alignment methods that align the distribution through adversarial training, which include AdaptSeg [54], CyCADA [26], CLAN [36], APODA [67], PatchAlign [56], ADVENT [57], BDL [34] and FADA [60]; 2) self-training approaches, including CBST”
What Can Be Transferred: Unsupervised Domain Adaptation for Endoscopic Lesions Segmentation (2020, background): “Metrics BL [9] LtA [37] CGAN [23] CLAN [29] ADV [38] BDL [26] SWES [16] DPR [36] PyCDA [27] Ours”
MetaCorrection: Domain-aware Meta Loss Correction for Unsupervised Domain Adaptation in Semantic Segmentation (2021, neutral): “Compared with domain alignment methods [40, 41, 20], the proposed method shows superior performance.”
Instance Adaptive Self-Training for Unsupervised Domain Adaptation (2020, background): “3 PatchAlign [28] 92.”
BAPA-Net: Boundary Adaptation and Prototype Alignment for Cross-domain Semantic Segmentation (2021, methodology): “Following the common experimental setup for crossdomain semantic segmentation [50, 37, 38, 51, 52, 61, 66, 64], we evaluate our method on two synthetic-to-real scenarios, namely GTA5 to Cityscapes and SYNTHIA to Cityscapes.”
FSDR: Frequency Space Domain Randomization for Domain Generalization (2021, background): “The first is adversarial training based which employs adversarial learning to align source and target distributions in the feature, output or latent space [22, 39, 69, 40, 66, 7, 81, 54, 55, 72, 1, 68, 26, 18, 25, 31].”
DACS: Domain Adaptation via Cross-domain Mixed Sampling (2020, methodology): “In particular, alignment on the semantic level has been explored with similar methods for both UDA [14] and SSL [4].”