Max-margin Latent Dirichlet Allocation for Image Classification and Annotation

Yang, Weijia; Mori, Greg

doi:10.5244/c.25.112

Cited by 29 publications

(39 citation statements)

References 18 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…We performed extensive quantitative comparisons of SupDoc-NADE with the original DocNADE model and supervised LDA (sLDA) 1 [10,9]. We also provide some comparisons with MMLDA [12] and a Spatial Pyramid Matching (SPM) approach [17]. The code to download the datasets and for SupDocNADE is available at https://sites.…”

Section: Experiments and Resultsmentioning

confidence: 99%

“…We also compare with MMLDA [12], which has been applied to image classification and annotation separately. The reported classification accuracy for MMLDA is less than SupDocNADE as shown in Table 1.…”

Section: Quantitative Comparisonmentioning

confidence: 99%

“…Wang et al [9] proposed a supervised LDA formulation to tackle this problem. Wang et al [12] opted instead for a maximum margin formulation of LDA (MMLDA). Our work also belongs to this line of work, extending topic models to a supervised variant: our contribution is thus to extend a different topic model, DocNADE, to this context for multimodal data modeling.…”

Section: Related Workmentioning

confidence: 99%

See 2 more Smart Citations

Topic Modeling of Multimodal Data: An Autoregressive Approach

Zheng

Zhang

Larochelle

2014

2014 IEEE Conference on Computer Vision and Pattern Recognition

View full text Add to dashboard Cite

Topic modeling based on latent Dirichlet allocation (LDA) has been a framework of choice to deal with multimodal data, such as in image annotation tasks. Recently, a new type of topic model called the Document Neural Autoregressive Distribution Estimator (DocNADE) was proposed and demonstrated state-of-the-art performance for text document modeling. In this work, we show how to successfully apply and extend this model to multimodal data, such as simultaneous image classification and annotation. Specifically, we propose SupDocNADE, a supervised extension of DocNADE, that increases the discriminative power of the hidden topic features by incorporating label information into the training objective of the model and show how to employ SupDocNADE to learn a joint representation from image visual words, annotation words and class label information. We also describe how to leverage information about the spatial position of the visual words for SupDocNADE to achieve better performance in a simple, yet effective manner. We test our model on the LabelMe and UIUC-Sports datasets and show that it compares favorably to other topic models such as the supervised variant of LDA and a Spatial Matching Pyramid (SPM) approach.

show abstract

Section: Experiments and Resultsmentioning

confidence: 99%

Section: Quantitative Comparisonmentioning

confidence: 99%

Section: Related Workmentioning

confidence: 99%

See 1 more Smart Citation

Topic Modeling of Multimodal Data: An Autoregressive Approach

Zheng

Zhang

Larochelle

2014

2014 IEEE Conference on Computer Vision and Pattern Recognition

View full text Add to dashboard Cite

show abstract

“…Meanwhile the annotation accuracy of our M 3 DA-RF model tends to a relatively stable value with the growing number of topics, which is mainly because the word-level features (also known as the soft label probability P SVM ) play a dominant role in the multi-level maxmargin discriminative feature space and the topic-level features are just helpful supplements. The most related work to ours is detailed in [16], where the authors only exploit different types of feature representation and do not make full use of the contextual information that may be beneficial to annotation tasks. Some other related studies [1,5,9] have investigated the application of topic model in satellite images annotation task.…”

Section: Discussionmentioning

confidence: 99%

Tile-Level Annotation of Satellite Images Using Multi-Level Max-Margin Discriminative Random Field

Yang

Chen

2013

Remote Sensing

View full text Add to dashboard Cite

This paper proposes a multi-level max-margin discriminative analysis (M 3 DA) framework, which takes both coarse and fine semantics into consideration, for the annotation of high-resolution satellite images. In order to generate more discriminative topic-level features, the M 3 DA uses the maximum entropy discrimination latent Dirichlet Allocation (MedLDA) model. Moreover, for improving the spatial coherence of visual words neglected by M 3 DA, conditional random field (CRF) is employed to optimize the soft label field composed of multiple label posteriors. The framework of M 3 DA enables one to combine word-level features (generated by support vector machines) and topic-level features (generated by MedLDA) via the bag-of-words representation. The experimental results on high-resolution satellite images have demonstrated that, using the proposed method can not only obtain suitable semantic interpretation, but also improve the annotation performance by taking into account the multi-level semantics and the contextual information.

show abstract

“…A great deal of research efforts have been devoted to bridge the semantic gap. Both low level visual features and high level semantics are explored in previous literatures [1][2][3][4][5][6][7][8][9][10].…”

Section: Introductionmentioning

confidence: 99%

Labeling Images by Integrating Sparse Multiple Distance Learning and Semantic Context Modeling

Zhou

Lin

et al. 2012

Computer Vision – ECCV 2012

View full text Add to dashboard Cite

Abstract. Recent progress on Automatic Image Annotation (AIA) is achieved by either exploiting low level visual features or high level semantic context. Integrating these two paradigms to further leverage the performance of AIA is promising. However, very few previous works have studied this issue in a unified framework. In this paper, we propose a unified model based on Conditional Random Fields (CRF), which establishes tight interaction between visual features and semantic context. In particular, Kernelized Logistic Regression (KLR) with multiple visual distance learning is embedded into the CRF framework. We introduce L1 and L2 regularization terms into the unified learning process for the distance learning and the parameters penalty respectively. The experiments are conducted on two benchmarks: Corel and TRECVID-2005 data sets for evaluation. The experimental results show that, compared with the state-of-the-art methods, the unified model achieves significant improvement on annotation performance and shows more robustness with increasing number of various visual features.

show abstract

Max-margin Latent Dirichlet Allocation for Image Classification and Annotation

Cited by 29 publications

References 18 publications

Topic Modeling of Multimodal Data: An Autoregressive Approach

Topic Modeling of Multimodal Data: An Autoregressive Approach

Tile-Level Annotation of Satellite Images Using Multi-Level Max-Margin Discriminative Random Field

Labeling Images by Integrating Sparse Multiple Distance Learning and Semantic Context Modeling

Contact Info

Product

Resources

About