Fast Approximate GMM Soft-Assign for Fine-Grained Image Classification with Large Fisher Vectors

Krapac, Josip; Šegvić, Siniša

doi:10.1007/978-3-319-24947-6_39

Cited by 2 publications

(4 citation statements)

References 15 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…This discrepancy is due to soft-assign stage being 33 times slower in the case of traffic signs. Our detectors may be accellerated further by using the fast approximation of the GMM soft-assignment, as proposed in [36]. We leave this, however, for future work.…”

Section: Discussionmentioning

confidence: 99%

“…The complexity of patch scoring can be subdivided into the following three stages with similar computational complexity: i) computing the soft-assign p(k|x), ii) computing the FV, and iii) determining the patch contribution by (6), (7), or (9). In this paper, we consider efficient implementation of the latter two stages for CPU architectures (efficient soft-assign is addressed in [36]). In the naive implementation, both of these two stages are O(NKD) where N is the number of patches, K is the number of components, while D is the raw feature dimensionality.…”

Section: Efficient Patch Scoring With a Sparse Modelmentioning

confidence: 99%

See 1 more Smart Citation

Sparse weakly supervised models for object localization in road environment

Zadrija¹,

Krapac

Šegvić

et al. 2018

Computer Vision and Image Understanding

Self Cite

View full text Add to dashboard Cite

We propose a novel weakly supervised localization method based on Fisher-embedding of low-level features (CNN, SIFT), and model sparsity at the component level. Fisher-embedding provides an interesting alternative to raw lowlevel features, since it allows fast and accurate scoring of image subwindows with a model trained on entire images. Model sparsity reduces overfitting and enables fast evaluation. We also propose two new techniques for improving performance when our method is combined with nonlinear normalizations of the aggregated Fisher representation of the image. These techniques are i) intra-component metric normalization and ii) first-order approximation to the score of a normalized image representation. We evaluate our weakly supervised localization method on real traffic scenes acquired from driver's perspective. The method dramatically improves the localization AP over the dense nonnormalized Fisher vector baseline (16 percentage points for zebra crossings, 21 percentage points for traffic signs) and leads to a huge gain in execution speed (91× for zebra crossings, 74× for traffic signs).

show abstract

Section: Discussionmentioning

confidence: 99%

Section: Efficient Patch Scoring With a Sparse Modelmentioning

confidence: 99%

Sparse weakly supervised models for object localization in road environment

Zadrija¹,

Krapac

Šegvić

et al. 2018

Computer Vision and Image Understanding

Self Cite

View full text Add to dashboard Cite

show abstract

“…We do not match the cascading approach of [36] or [12] for a single class scenario, but our approach might scale better in the multi-class case because different classes may share features through a shared GMM. We could further speed-up the process by using fast soft assign as proposed in [20] or by using the random decision forests as a generative model for FV [3]. An additional speed-up could be achieved by using the C implementation instead of Python.…”

Section: Methodsmentioning

confidence: 99%

“…Due to generative front-end, we have a better sharing potential than pure discriminative approaches used in [27]. In contrast with [21], we use block-sparsity [20], the normalized score gradient and the spatial model of the pairwise layout.…”

Section: Related Workmentioning

confidence: 99%

Patch-Level Spatial Layout for Classification and Weakly Supervised Localization

Zadrija

Krapac

Verbeek

et al. 2015

Lecture Notes in Computer Science

Self Cite

View full text Add to dashboard Cite

Abstract. We propose a discriminative patch-level spatial layout model suitable for training with weak supervision. We start from a block-sparse model of patch appearance based on the normalized Fisher vector representation. The appearance model is responsible for i) selecting a discriminative subset of visual words, and ii) identifying distinctive patches assigned to the selected subset. These patches are further filtered by a sparse spatial model operating on a novel representation of pairwise patch layout. We have evaluated the proposed pipeline in image classification and weakly supervised localization experiments on a public traffic sign dataset. The results show significant advantage of the proposed spatial model over state of the art appearance models.

show abstract

Fast Approximate GMM Soft-Assign for Fine-Grained Image Classification with Large Fisher Vectors

Cited by 2 publications

References 15 publications

Sparse weakly supervised models for object localization in road environment

Sparse weakly supervised models for object localization in road environment

Patch-Level Spatial Layout for Classification and Weakly Supervised Localization

Contact Info

Product

Resources

About