Learning a Fine Vocabulary

By Andrej Mikulík, Michal Perdoch, Ondřej Chum, Jiri Matas. 2010

159 citations recorded by the source corpus.

Subfields: Computer Vision and Pattern Recognition, Natural Language Processing

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Computer Vision and Pattern Recognition

In Natural Language Processing

Citations in the corpus

PlaNet - Photo Geolocation with Convolutional Neural Networks (2016, methodology): “igure 9. Top: GeoGuessr panorama, Bottom: Ground truth location (yellow), human guess (green), PlaNet guess (blue). Method Holidays mAP BoVW 57.2 [24] Hamming Embedding 77.5 [24] Fine Vocabulary 74.9 [34] ASMK+MA 82.2 [47] GIST 37.6 [10] Overfeat 64.2 [39] Overfeat+aug+ss 84.3 [39] AlexNet+LM Retrai”
CNN Features Off-the-Shelf: An Astounding Baseline for Recognition (2014, methodology): “[26] use in order of million codebooks but with various tricks including bi-”
NetVLAD: CNN Architecture for Weakly Supervised Place Recognition (2015, background): “Numerous works concentrate on learning better local descriptors or metrics to compare them [45, 48, 50, 55, 56, 69, 70, 86], but even though some of them show results on image retrieval, the descriptors are learnt on the task of matching local image patches, and not directly with image retrieval in ”
End-to-End Learning of Deep Visual Representations for Image Retrieval (2016, background): “C V ] 2 5 O ct 2 tial verification (Mikuĺık et al, 2010; Tolias et al, 2015; Tolias and Jégou, 2015; Li et al, 2015).”
Visual Place Recognition with Repetitive Structures (2015, background): “Examples include: (i) learning better visual vocabularies [21, 28]; (ii) developing quantization methods less prone to quantization errors [14, 27, 44]; (iii) combining returns from multiple query images depicting the same scene [4, 6]; (iv) exploiting the 3D or graph structure of the database [11, ”
Leveraging Structure from Motion to Learn Discriminative Codebooks for Scalable Landmark Classification (2013, methodology): “Large codebooks learned in an unsupervised setting can also be discriminative [22, 26], but they are only suitable for large-scale image retrieval [22, 26, 20].”
SIFT Meets CNN: A Decade Survey of Instance Retrieval (2016, methodology): “Most SIFT-based methods employ the original images, except [32], [71] which manually rotate the images into upright orientations.”
Three things everyone should know to improve object retrieval (2012, methodology): “[17], (j), where a very large vocabulary is generated and similarities between the words are learnt.”
Searching in one billion vectors: Re-rank with source coding (2011, background): “Testing on a large scale is important, as most ANN methods are usually evaluated on sets of unrealistic size, thereby ignoring memory issues that arise in real applications, where billions of vectors have to be handled [4, 9].”
Hello neighbor: Accurate object retrieval with k-reciprocal nearest neighbors (2011, background): “Descriptor space learning techniques [19, 15] push relevant images closer to the query by correcting for the fact that the Euclidean norm is not a perfect distance measure in SIFT or SURF descriptor spaces.”
Visual Place Recognition: A Survey (2016, background): “[211] learned a fine vocabulary [212]; a fine vocabulary is similar to a bag-of-words model in that it”
Learning and Calibrating Per-Location Classifiers for Visual Place Recognition (2013, background): “Examples include: (i) learning better visual vocabularies from training examples with matched/non-matched descriptors [19, 23]; (ii) developing quantization methods less prone to quantization er-”
Graph-Based Discriminative Learning for Location Recognition (2013, methodology): “use image matches to compute global correlations between visual words [21].”
To Aggregate or Not to aggregate: Selective Match Kernels for Image Search (2013, unknown): “In a query by content scenario [25], which is the focus of our paper, large vocabularies make the search efficient [17, 22, 16], thanks to inverted file structures [24] that exploit the sparsity of the representation.”
Total recall II: Query expansion revisited (2011, background): “Virtually all aspects of particular object BoW-type retrieval have been intensively studied: feature detectors and descriptors [14, 5, 25, 15, 16], vocabulary construction [23, 18, 21, 9, 17], spatial verification and re-ranking [21, 9], document metric learning [12, 10, 6] and dimensionality reduct”
Learning visual feature descriptors for dynamic lighting conditions (2014, methodology): “[10] use the fine vocabulary method of [21] to learn a probability distribution over visual words in an attempt to capture which visual words can be produced by the same scene feature under various lighting conditions.”
Cross-Dimensional Weighting for Aggregated Deep Convolutional Features (2015, background): “When further combining our approach with query expansion, we get even better results that compare to (or surpass on Paris) far more sophisticated approaches like [6,22,34] that are based on local features and include spatial verification steps.”
Light-weight localization for vehicles using road markings (2013, background): “Approaches to reliable localization over long periods of time include the use of high resolution panoramic images [29], learning appearance change in feature descriptors [15][21], and creating a new map whenever localization fails [3].”
Deep Image Retrieval: Learning Global Representations for Image Search (2016, unknown): “In fact, for most image retrieval benchmarks, the state of the art is currently held by conventional methods relying on local descriptor matching and re-ranking with elaborate spatial verification [8,9,10,11].”