Kernel Descriptors for Visual Recognition

By Liefeng Bo, Xiaofeng Ren, Dieter Fox. 2010

375 citations recorded by the source corpus.

Subfields: Computer Vision and Pattern Recognition

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

Citations in the corpus

DeCAF: A Deep Convolutional Activation Feature for Generic Visual Recognition (2013, methodology): “As our ﬁrst approach, we resized each predicted part box to 256 × 256 and computed DeCAF 6 to replace KDES image features (Bo et al., 2010) used in DPD paper.”
A Joint Model of Language and Perception for Grounded Attribute Learning (2012, background): “Recent work on kernel descriptors (Bo et al., 2010) shows that these hand-designed features are equivalent to a type of match kernel that performs similarly to sparse coding (Yang et al., 2009; Yu & Zhang, 2010) and deep networks (Lee et al., 2009) on many object recognition benchmarks (Bo et al.,…”
The application of two-level attention models in deep convolutional neural network for fine-grained image classification (2014, methodology): “We compare against two baseline feature extractors, one is hand-crafted kernel descriptors [3] (KDES) which was widely used in fine-grained classification before using CNN feature, the other is the CNN feature extractor pre-trained from all the data in ILSVRC2012 [16].”
Convolutional Kernel Networks (2014, methodology): “The convolutional multilayer kernel is a generalization of the hierarchical kernel descriptors introduced in computer vision [2, 3].”
Bird Species Categorization Using Pose Normalized Deep Convolutional Nets (2014, methodology): “Recently methods that employed more modern features like POOF [5], Fisher-encoded SIFT and color descriptors [40], and Kernel Descriptors (KDES) [7] significantly boosted performance into the 50− 62% range [5, 12, 15, 21, 51].”
Hand-Crafted Features or Machine Learnt Features? Together They Improve RGB-D Object Recognition (2014, methodology): “Kernel descriptors [3] measure similarities between image patches and learn compact patch-level features using pixel attributes (gradient, color, local binary pattern, etc.”
Local Group Invariant Representations via Orbit Embeddings (2016, methodology): “A kernel view of histogram of gradients was introduced in [5], where finite dimensional features were defined through kernel PCA.”
Which Looks Like Which: Exploring Inter-class Relationships in Fine-Grained Visual Categorization (2014, methodology): “We compare with the following published approaches: multiple kernel learning (MKL) [11], Birdlet [19], a random template method [41], and the KDES [10].”
Pairwise Rotation Invariant Co-Occurrence Local Binary Pattern (2014, background): “…role in a wide range of applications that includes static and dynamic texture classification [1], [2], [3], [4], [5], object and scene recognition [6], [7], [8], [9], [10], face detection and recognition [11], image retrieval [12], [13], stereo correspondence, 3D reconstruction and many more.”
Local Alignments for Fine-Grained Categorization (2014, methodology): “…parts (using pooling-based encodings), rather than using shape - sensitive descriptors such as HOG (Dalal and Triggs 2005) or kernel descriptors (Bo et al. 2010) on deformable parts (Felzenszwalb et al. 2010) as used in (Farrell et al. 2011; Yang et al. 2012; Chai et al. 2013; Zhang et al.…”
Multipath Sparse Coding Using Hierarchical Matching Pursuit (2013, background): “While SIFT can be understood and generalized as a way to go from pixels to patch descriptors [2], designing good features is a challenging task that requires deep domain knowledge, and it is often difficult to adapt to new settings.”
Augmenting Strong Supervision Using Web Data for Fine-Grained Categorization (2015, background): “The second strategy is to introduce more discriminative feature representations [5, 4, 34], which is particularly attractive given by the recent success of deep convolutional neural networks (CNNs) [19, 11] in visual recognition.”
Unsupervised Template Learning for Fine-Grained Object Recognition (2012, methodology): “Methods SIFT [27] KDES [3] Edge templates Texture templates Accuracy(%) 22.”
Deformable Part Descriptors for Fine-Grained Recognition and Attribute Prediction (2013, methodology): “Given the pose-normalized representation Ψpn(I) in Equation 3, we employ a linear SVM for the final classification.”
Discriminative Learning of Sum-Product Networks (2012, neutral): “5% SIFT [5] 65.”
PCANet: A Simple Deep Learning Baseline for Image Classification? (2014, neutral): “Methods Accuracy Tiled CNN [54] 73.10 Improved LCC [55] 74.50 KDES-A [56] 76.00 K-means (Triangle, 4000 features) [57] 79.60 Cuda-convnet2 [58] 82.00 Stochastic pooling ConvNet [44] 84.87 CNN + Spearmint [59] 85.02 Conv.”
Learning Semantic Image Representations at a Large Scale (2014, methodology): “As our first approach, we resized each predicted part box to 256 × 256 and computed FC6 to replace the KDES image features [8] used in the original DPD paper [124].”
On the Similarity between the Laplace and Neural Tangent Kernels (2020, methodology): “For our final experiment we constructed alternative hierarchical kernels, in the spirit of [12, 33], by recursively applying exponential kernels in a manner similar to CNTK.”