The devil is in the details: an evaluation of recent feature encoding methods

By Ken Chatfield, Victor S. Lempitsky, Andrea Vedaldi, Andrew Zisserman. 2011

953 citations recorded by the source corpus.

Subfields: Computer Vision and Pattern Recognition, Media Technology, Representation & Self-Supervised Learning, Evaluation & Benchmarks

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

In Representation & Self-Supervised Learning

In Evaluation & Benchmarks

Citations in the corpus

Deep Residual Learning for Image Recognition (2015, background): “with respect to a dictionary, and Fisher Vector [30] can be formulated as a probabilistic version [18] of VLAD. Both of them are powerful shallow representations for image retrieval and classiﬁcation [4,48]. For vector quantization, encoding residual vectors [17] is shown to be more effective than e”
Spatial Pyramid Pooling in Deep Convolutional Networks for Visual Recognition (2014, methodology): “Analogously, the deep convolutional features can be pooled in a similar way.”
Learning Visual Features from Large Weakly Supervised Data (2015, background): “We note that achieving state-of-the-art results [5, 34, 39, 60] on these datasets requires the development of tailored pipelines, e.g. , using many image transformations and model ensembles, which is out of the scope of this paper.”
Fine-Grained Visual Classification of Aircraft (2013, background): “shows the confusion matrix for a strong baseline model (non-linear SVM on a ˜2 kernel, bag-of-visual words, 600 k-means words dictionary, multi-scale dense SIFT features, and 1 1, 2 2 spatial pyramid [1]). These models were trained on the entire image ignoring the bounding box information. As seen i”
Unmasking Clever Hans predictors and assessing what machines really learn (2019, methodology): “1 The Fisher Vector Classifier The FV model evaluated in this section follows the configuration and implementation from [31, 142] and can easily be replicated by using the Encoding Evaluation Toolkit [142].”
Fisher Vectors Derived from Hybrid Gaussian-Laplacian Mixture Models for Image Annotation (2014, positive): “However, the leading pooling technique in recent years is the Fisher Vector [32], which provided state-of-the-art results on many different applications [38, 31, 2, 33].”
Bag of visual words and fusion methods for action recognition: Comprehensive study and good practice (2014, neutral): “Keywords Action recognition · Bag of Visual Words · Fusion methods · Survey”
P-CNN: Pose-Based CNN Features for Action Recognition (2015, methodology): “Fisher Vectors (FV) [27] encoding has been shown to outperform the bag-of-word approach [5] resulting in stateof-the-art performance for action recognition in combination with DT features [25].”
CRAFT Objects from Images (2016, methodology): “two popular models are the Deformable Part Model (DPM [9]) and the Bag of Words (BOW [25, 3]).”
Analyzing Classifiers: Fisher Vectors and Deep Neural Networks (2015, background): “1) from [3] with settings as in this paper.”
Label-Embedding for Image Classification (2015, positive): “These descriptors are, then, aggregated into an image-level representation using the Fisher Vector (FV) [45], shown to be a state-of-the-art patch encoding technique in [8].”
Sympathy for the Details: Dense Trajectories and Hybrid Classification Architectures for Action Recognition (2016, methodology): “ multiple trajectory descriptors in a single video-level representation by accumulating their Fisher Vector encodings (FV) [24,25], which was shown to be particularly effective for action recognition [5,46]. This high-dimensional representation is based on the gradient of a generative model, a Gauss”
Good Practice in Large-Scale Learning for Image Classification (2014, methodology): “[16] benchmarked five local descriptor encoding techniques (BOV with hard coding, soft coding and sparse coding as well as SV and FV coding); the FV of [57] was shown to yield the best results on two standard data sets.”
LEAR-INRIA submission for the THUMOS workshop (2013, background): “In recent evaluations [2, 7], this shows an improved performance over bag of features for both image and action classification.”