Unsupervised Visual Domain Adaptation Using Subspace Alignment

Fernando, Basura; Habrard, Amaury; Sebban, Marc; Tuytelaars, Tinne

doi:10.1109/iccv.2013.368

Cited by 1,111 publications

(967 citation statements)

References 10 publications

Supporting

Mentioning

962

Contrasting

Unclassified

Order By: Relevance

“…Successful applications in vision include knowledge transfer from one data source to another (e.g. videos to images) [17,23,32] and knowledge transfer from known classes to unseen classes [33]. The difference between such work and ours is in the form of the knowledge.…”

Section: Transfer Learning and Domain Adaptationmentioning

confidence: 96%

Metric imitation by manifold transfer for efficient vision applications

Dai

Kroeger

Timofte

et al. 2015

2015 IEEE Conference on Computer Vision and Pattern Recognition (CVPR)

View full text Add to dashboard Cite

Metric learning has proved very successful. However, human annotations are necessary. In this paper, we propose an unsupervised method, dubbed Metric Imitation (MI), where metrics over cheap features (target features, TFs) are learned by imitating the standard metrics over more sophisticated, off-the-shelf features (source features, SFs) by transferring view-independent property manifold structures. In particular, MI consists of: 1) quantifying the properties of source metrics as manifold geometry, 2) transferring the manifold from source domain to target domain, and 3) learning a mapping of TFs so that the manifold is approximated as well as possible in the mapped feature domain. MI is useful in at least two scenarios where: 1) TFs are more efficient computationally and in terms of memory than SFs; and 2) SFs contain privileged information, but are not available during testing. For the former, MI is evaluated on image clustering, category-based image retrieval, and instance-based object retrieval, with three SFs and three TFs. For the latter, MI is tested on the task of example-based image super-resolution, where high-resolution patches are taken as SFs and low-resolution patches as TFs. Experiments show that MI is able to provide good metrics while avoiding expensive data labeling efforts and that it achieves state-of-the-art performance for image super-resolution. In addition, manifold transfer is an interesting direction of transfer learning.

show abstract

Section: Transfer Learning and Domain Adaptationmentioning

confidence: 96%

Metric imitation by manifold transfer for efficient vision applications

Dai

Kroeger

Timofte

et al. 2015

2015 IEEE Conference on Computer Vision and Pattern Recognition (CVPR)

View full text Add to dashboard Cite

show abstract

“…For subspace alignment (SA), the first step is to generate the subspace for both source and target data [5], which is realized by principal component analysis (PCA). Then, an alignment between those subspaces is learned.…”

Section: Subspace Alignmentmentioning

confidence: 99%

Investigation of domain adaptation for acoustic frog species classification

Xie

2018

Preprint

View full text Add to dashboard Cite

Acoustic frog species classification has received much attention for its importance in assessing biodiversity. However, most previous frog call classification models are trained and tested using the data collected from the same area, which greatly limits the model's generalization. In practice, frogs often have regional accents. When training and testing data are collected from different areas, there is an adverse impact on frog call classification performance. To tackle this problem, this paper investigates domain adaptation for classifying frog calls collected from different areas. To evaluate the performance of our proposed methods, two frog call datasets, which are collected from subtropical eastern Australia and tropical north-eastern Australia, are used. Experimental results demonstrate that domain adaptation can significantly improve the weighted F1-score from 72.8% to 85.5%.

show abstract

“…It is closely related to inductive transfer, but we usually find that many tasks are simultaneously learnt. In other words, there is a generic model that is built from several tasks, usually by transforming or augmenting the feature space into a more general representation (e.g., using subspaces described by eigenvectors [20]), or by coding general and targetspecific versions of the features [12]. However, despite its apparent similarity with input reframing, and the extraction of invariants across domains or the use of common parameters across domains [14], these do not really represent a context.…”

Section: Reframingmentioning

confidence: 99%