In defense of Nearest-Neighbor based image classification

By Oren Boiman, Eli Shechtman, Michal Irani. 2008 IEEE Conference on Computer Vision and Pattern Recognition, 2008

1,301 citations recorded by the source corpus.

Subfields: Computer Vision and Pattern Recognition, Media Technology

View Paper: https://www.semanticscholar.org/paper/7ae4d5ec354f2a54b5cd70395cb12283390d0638

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 Media Technology

Citations in the corpus

Revisiting Local Descriptor Based Image-To-Class Measure for Few-Shot Learning (2019, methodology): “As indicated in [1], this setting can be justiﬁed by a fact that a new image can be roughly “composed” by using the pieces of other images in the same class ( i.e. , the exchangeability of visual patterns across the images in the same class).”
Optimized Product Quantization for Approximate Nearest Neighbor Search (2013, background): “Approximate nearest neighbor (ANN) search is of great importance for many computer vision problems, such as retrieval [17], classification [2], and recognition [18].”
Visual Word Ambiguity (2010, methodology): “The most frequent features, as noted by Jurie and Triggs [12] and others [3], [23], are not necessarily the most discriminative.”
Multi-spectral SIFT for scene category recognition (2011, methodology): “We perform scene recognition on our dataset of 477 images, randomly selecting 99 images for testing (11 per category) and training using the rest.”
Detection of Co-salient Objects by Looking Deep and Wide (2016, methodology): “(7) to (9) are inspired by the negative mining technology used in Siva et al. (2012, 2013), Boiman et al. (2008).”
Linear spatial pyramid matching using sparse coding for image classification (2009, methodology): “Algorithms 15 training 30 training Zhang et al. [28] 59.10 ± 0.60 66.20 ± 0.50 KSPM [12] 56.40 64.40 ± 0.80 NBNN [1] 65.00 ± 1.14 70.40 ML+CORR [9] 61.00 69.60 KC [25] – 64.14 ± 1.18 KSPM 56.44 ± 0.78 63.99 ± 0.88 LSPM 53.23 ± 0.65 58.81 ± 1.51 ScSPM 67.0 ± 0.45 73.2 ± 0.54”
A Review on Human Activity Recognition Using Vision-Based Method (2017, methodology): “Also, a most recent evaluation [58] showed a consistent ﬁ nding on UCF-YouTube and HMDB51 datasets, though slightly slower than local NBNN on KTH.”
High-level event recognition in unconstrained videos (2012, methodology): “Besides, these methods also suffer from the usual disadvantages of quantization used in converting raw features to discrete codewords as pointed out in [16,109].”
K-Means Hashing: An Affinity-Preserving Quantization Method for Learning Binary Compact Codes (2013, background): “Approximate nearest neighbors (ANN) search is widely applied in image/video retrieval [27, 11], recognition [28], image classification [23, 2], pose estimation [25], and many other computer vision problems.”
Retrieval of Brain Tumors by Adaptive Spatial Pooling and Fisher Vector Representation (2016, methodology): “For instance, the scale-invariant feature transform (SIFT) descriptor, a well-known local feature that is typically extracted from key points, has demonstrated its excellent robustness and discriminative power in natural image classification and retrieval tasks [15,19 – 21].”
Learning mid-level features for recognition (2010, neutral): “[3] Nearest neighbor + spatial correspondence 65.”
Proposal : Scalable , Active and Flexible Learning on Distributions (2015, neutral): “Some distributional learning methods do not directly compare sample sets to one another, but rather compare their elements to a class-level distribution (Boiman et al. 2008).”
Geometric Neural Phrase Pooling: Modeling the Spatial Co-occurrence of Neurons (2016, methodology): “These Image-level vectors are then normalized and fed into machine learning algorithms [21] for training and testing, or used in some training-free image classification algorithms [7,22].”
ImageNet: A large-scale hierarchical image database (2009, methodology): “Nearest Neighbor (NBNN) method proposed in [ 5 ] to underline the usefulness of full resolution images.”
Recognizing Actions from Depth Cameras as Weakly Aligned Multi-part Bag-of-Poses (2013, background): “In particular, the experiments in [4] showed that frequent descriptors have low quantization error, but rare descriptors have high quantization error.”
Improving patch-based scene text script identification with ensembles of conjoined networks (2016, background): “C V ] 1 F eb Keywords: script identification, scene text understanding, multi-language OCR, convolutional neural networks, ensemble of conjoined networks”