Negative evidences and co-occurrences in image retrieval: the benet of PCA and whitening

By Ondřej Chum. 2012

353 citations recorded by the source corpus.

View Paper: https://www.semanticscholar.org/paper/c154d4d80757f9e07248817861af72681dcba5b7

How researchers see this paper

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

Citations in the corpus

“Hands On” Visual Recognition for Visually Impaired Users (2017, methodology): “It is also worth highlighting the very good performance of BoF, MBoFQ, VLAD, and Fisher PCA-compressed vectors for cans recognition.”
Supervised mid-level features for word image representation (2014, methodology): “iii) Project the FVs with U, and `2 normalize again – `2 normalization after PCA/CCA usually improves the results since it accounts for the missing energy in the dimensionality reduction step [14].”
All About VLAD (2013, methodology): “614 Reimplemented Multivoc-VLAD [5] 0.”
Mining on Manifolds: Metric Learning Without Labels (2018, methodology): “We follow the common practice in the literature and perform unsupervised PCA whitening [23] for the pre-trained networks, and supervised LDA-based whitening for the fine-tuned networks, learned on a subset of Flickr 7M.”
3D Pose from Motion for Cross-View Action Recognition via Non-linear Circulant Temporal Encoding (2014, unknown): “However, this is not the case for the BoW that describe our frames.”
InLoc: Indoor Visual Localization with Dense Matching and View Synthesis (2019, background): “The impact of negative evidence in feature aggregation is demonstrated in [36].”
Grassmann Pooling as Compact Homogeneous Bilinear Pooling for Fine-Grained Visual Classification (2018, methodology): “VLAD and Fisher Vector.”
Multi-scale Orderless Pooling of Deep Convolutional Activation Features (2014, methodology): “Using aggressive dimensionality reduction with PCA and whitening as suggested in [39], we can raise the mAP even further to 80.”
Fine-Tuning CNN Image Retrieval with No Human Annotation (2017, methodology): “The PCA-whitening [22] (PCA w ) is shown to be essential in some cases of CNN-based descriptors [10], [12], [15].”
Learning With Average Precision: Training Image Retrieval With a Listwise Loss (2019, methodology): “As in [28], we use a square-rooted PCA.”
Fisher vectors meet Neural Networks: A hybrid classification architecture (2015, methodology): “PCA followed by whitening has been recently rediscovered as a useful transformation for tasks such as retrieval and classification [23] or detection”
Deep Adaptive Image Clustering (2017, background): “Many applications such as content-based image annotation [19, 20, 22, 23] and image retrieval [12, 24, 34] can be viewed as different instances of image clustering.”
Multi-view Super Vector for Action Recognition (2014, methodology): “We also whiten each types of descriptors as is suggested in [13].”
End-to-End Learning of Deep Visual Representations for Image Retrieval (2016, methodology): “G lo b a l d es cr ip to rs Jégou and Zisserman (2014) 1024 56.0 50.2 72.0 Jégou and Zisserman (2014) 128 43.3 35.3 61.7 Gordo et al (2012) 512 79.”
Visual Instance Retrieval with Deep Convolutional Networks (2014, methodology): “Popular approaches for this goal include vector aggregation methods such as the bag-of-words (BOW) representation (Sivic & Zisserman, 2003), the Fisher Vector (FV) (Perronnin & Dance, 2007), and the VLAD representation (Jégou et al., 2012); commonly in these methods each local image patch is first d”
Deep Image Retrieval: Learning Global Representations for Image Search (2016, methodology): “Encoding techniques, such as the Fisher Vector [24], or VLAD [25], combined with compression [26,27,28] produce global descriptors that scale to larger databases at the cost of reduced accuracy.”