AutoDIAL: Automatic Domain Alignment Layers

By Fabio Maria Carlucci, Lorenzo Porzi, Barbara Caputo, Elisa Ricci, Samuel Rota Bulò. 2017

335 citations recorded by the source corpus.

Classifiers trained on given databases perform poorly when tested on data acquired in different settings. This is explained in domain adaptation through a shift among distributions of the source and target domains. Attempts to align them have traditionally resulted in works reducing the domain shift by introducing appropriate loss terms, measuring the discrepancies between source and target distributions, in the objective function. Here we take a different route, proposing to align the learned representations by embedding in any given network specific Domain Alignment Layers, designed to match the source and target feature distributions to a reference one. Opposite to previous works which define a priori in which layers adaptation should be performed, our method is able to automatically learn the degree of feature alignment required at different levels of the deep network. Thorough experiments on different public benchmarks, in the unsupervised setting, confirm the power of our approach.

Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition

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

Citations in the corpus

Learning to cluster in order to Transfer across domains and tasks (2017, background): “…of work dealing with domain shift between image datasets by minimizing domain discrepancy (Tzeng et al., 2014; Long et al., 2015b; Ganin et al., 2016; Sun et al., 2016; Long et al., 2016; Sener et al., 2016; Carlucci et al., 2017; Long et al., 2017; Zellinger et al., 2017; Bousmalis et al., 2017).”
Domain Adaptive Faster R-CNN for Object Detection in the Wild (2018, background): “Recent works aim to improve the domain adaptability of deep neural networks, including [40, 15, 18, 50, 45, 43, 34, 24, 41, 42].”
Agnostic Domain Generalization (2018, background): “The idea is to introduce different measures of domain distribution shift at a single or multiple levels in the architecture [26, 41, 4, 5, 32] and then train the network to minimize these measures while also reducing a task-speciﬁc (e.g. classiﬁcation, detection) loss.”
Evaluating Prediction-Time Batch Normalization for Robustness under Covariate Shift (2020, background): “AutoDIAL [Cariucci et al., 2017] mixes data from the source and target domains during training time before passing them through the batch norm layers, whereas TransNorm [Wang et al., 2019] uses source and target domain batch norm statistics during training in an end-to-end fashion to improve…”
A Survey of Unsupervised Deep Domain Adaptation (2018, methodology): “AutoDIAL[27] N,TD ✓ not BN”
A comprehensive review of digital twin — part 1: modeling and twinning enabling technologies (2022, methodology): “One of the most obvious approaches to overcome the domain gap is distribution alignment approaches by statistics matching, including Adaptive Batch Normalization (AdaBN) and Automatic DomaIn Alignment Layers (AutoDIAL), using the statistics of source and target mini-batches at different layers to al”
Cross-Domain Few-Shot Classification via Learned Feature-Wise Transformation (2020, methodology): “In addition to image stylization and generation, conditional normalization has also been applied to align different data distributions for domain adaptation (Cariucci et al., 2017; Li et al., 2017b).”
VisDA: A Synthetic-to-Real Benchmark for Visual Domain Adaptation (2018, neutral): “COIL20 [11] 1,440 20 1 (tools) Office [16] 1,410 31 3 (office) Caltech [2] 1,123 10 1 (office) CAD-office [12] 775 20 1 (office) Cross-Dataset [22] 70,000+ 40 12 (mixed)”
AdaGraph: Unifying Predictive and Continuous Domain Adaptation Through Graphs (2019, methodology): “Following recent advances in DA [3, 20, 23], we propose to learn a set of domain-specific models by considering a common backbone network with domain-specific alignment layers embedded into it.”
Importance Weighted Adversarial Nets for Partial Domain Adaptation (2018, methodology): “Since recent advance in deep learning has shown that more transferable and domain invariant features can be extracted through deep framework, the domain adaptation techniques are also transferred from shallow learningbased [1, 20, 15, 16, 8, 6, 28] to deep learning-based [25, 14, 18, 17, 27, 22, 23,”
ROAD: Reality Oriented Adaptation for Semantic Segmentation of Urban Scenes (2017, background): “Recent works aim to improve the domain adaptability of deep neural networks, including [26, 8, 9, 36, 30, 29, 22, 14, 27, 28].”
Learning Semantic Representations for Unsupervised Domain Adaptation (2018, background): “, 2017b), Automatic Domain Alignment Layer (AutoDIAL) (Carlucci et al., 2017).”
Unbiased Mean Teacher for Cross-domain Object Detection (2020, background): “Many previous works aim to minimize the distance metric such as maximum mean discrepancy(MMD) [3, 30, 31, 32, 45].”
Bidirectional Learning for Domain Adaptation of Semantic Segmentation (2019, methodology): “As the extension to MMD, some statistics of feature distributions such as mean and covariance [2, 21] are used to match two different domains.”
Tent: Fully Test-Time Adaptation by Entropy Minimization (2021, methodology): “Entropy Minimization Entropy minimization is a key regularizer for domain adaptation (Carlucci et al., 2017; Shu et al., 2018; Saito et al., 2019; Roy et al., 2019), semi-supervised learning (Grandvalet & Bengio, 2005; Lee, 2013; Berthelot et al.”
Source Data-Absent Unsupervised Domain Adaptation Through Hypothesis Transfer and Labeling Transfer (2020, background): “ﬁers [51] or a semantic multi-output classiﬁer [52], [53] instead of a feature-conditional domain discriminator [42], respectively. There are also some other studies investigating batch normalization [54], [55] and adversarial dropout [56], [57] within the network architecture to ensure feature inva”
Domain-Specific Batch Normalization for Unsupervised Domain Adaptation (2019, background): “Domain alignment layers used in [15] deal with the domain shift problem by aligning source and target distributions to a reference Gaussian distribution.”
Semi-Supervised and Unsupervised Deep Visual Learning: A Survey (2022, methodology): “Co-training ((5)) is also introduced for UDA, which imputes training targets from multiple co-trained clas-siﬁers to learn from unlabeled data and match cross-domain distributions [209].”
EEG-Based Emotion Recognition Using Regularized Graph Neural Networks (2019, background): “[50] performed domain adaptation by modulating the batch norm layers’ statistics from source to target domain.”