Cross-Dimensional Weighting for Aggregated Deep Convolutional Features

By Yannis Kalantidis, Clayton Mellina, Simon Osindero. ECCV Workshops, 2015

455 citations recorded by the source corpus.

We propose a simple and straightforward way of creating powerful image representations via cross-dimensional weighting and aggregation of deep convolutional neural network layer outputs. We first present a generalized framework that encompasses a broad family of approaches and includes cross-dimensional pooling and weighting steps. We then propose specific non-parametric schemes for both spatial- and channel-wise weighting that boost the effect of highly active spatial responses and at the same time regulate burstiness effects. We experiment on different public datasets for image search and show that our approach outperforms the current state-of-the-art for approaches based on pre-trained networks. We also provide an easy-to-use, open source implementation that reproduces our results.

Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition

View Paper: https://www.semanticscholar.org/paper/dc3735c34797553aee3c100db654b70c0c819a89

How researchers see this paper

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

In Artificial Intelligence

In Computer Vision and Pattern Recognition

Citations in the corpus

NetVLAD: CNN Architecture for Weakly Supervised Place Recognition (2015, background): “While there have been many improvements in designing better image retrieval [22], [23], [24], [25], [48], [49], [50], [51], [52], [53], [54], [55], [56], [57], [58], [59], [60], [61], [62] and place recognition [2], [3], [4], [5], [6], [7], [8], [9], [10], [11], [12], [13], [14], [15], [16] systems,”
Large-Scale Image Retrieval with Attentive Deep Local Features (2016, background): “Other works have re-evaluated and proposed different feature aggregation methods using such deep local features [3, 21].”
Selective Convolutional Descriptor Aggregation for Fine-Grained Image Retrieval (2016, methodology): “crease. Because our method has less dimensions and is simple to implement, SCDA is more scalable for large-scale datasets without strong annotations and is easier to generalize. In addition, the CroW [21] paper presented the classication accuracy on CUB200-2011 without any netuning (56.5% by VGG-16)”
Deep Feature Factorization for Content-Based Image Retrieval and Localization (2019, methodology): “To begin, we adopt a method proposed by [6], known as CroW.”
A Discriminatively Learned CNN Embedding for Person Reidentification (2016, unknown): “Method mAP mVoc/BoW[16] 0.488 Neural codes(A)[2] 0.557 R-MAC(A)[20] 0.561 R-MAC-Hard(A)[17] 0.625 Ours(A) 0.662 CroW(V) [9] 0.654 R-MAC(V)[20] 0.669 R-MAC-Hard(V)[17] 0.770 MAC-Hard(V)[17] 0.797 Ours(V) 0.764 Table 7.”
Revisiting Oxford and Paris: Large-Scale Image Retrieval Benchmarking (2018, methodology): “Numerous methods of image retrieval [7, 31, 5, 27, 47, 3, 48, 20, 37, 10] and visual localization [9, 1] have used these datasets for evaluation.”
Sipakmed: A New Dataset for Feature and Image Based Classification of Normal and Pathological Cervical Cells in Pap Smear Images (2018, methodology): “Following the trend of a number of recent works, we employ the network as a feature extractor [8, 9, 10, 11, 12, 13].”
MetaSearch: Incremental Product Search via Deep Meta-Learning (2020, methodology): “Compared with the original CroW [27], FCroW considers both the nonzero and zero parts of the feature map to discover the salient channels.”
Learned Contextual Feature Reweighting for Image Geo-Localization (2017, methodology): “To demonstrate the benefit of our learnt CRN, we also compared our method with Crow [26] which performs feature reweighting in a predefined way.”
Fine-Tuning CNN Image Retrieval with No Human Annotation (2017, methodology): “It has recently become a standard policy to combine CNN global image descriptors with simple average query expansion (AQE) [10], [11], [12], [27].”
Particular object retrieval with integral max-pooling of CNN activations (2015, background): “Finally, Kalantidis et al. (2015) propose spatial and feature channel weighting that signiﬁcantly improves performance.”
CNN Image Retrieval Learns from BoW: Unsupervised Fine-Tuning with Hard Examples (2016, methodology): “[22] concurrently propose the use of Fully Connected (FC) layer activations as descriptors, while convolutional layer activations are later shown to have superior performance [15,23,24,25].”
SIFT Meets CNN: A Decade Survey of Instance Retrieval (2016, methodology): “For CNN-based methods, some employ the full-sized query images [8], [137]; some follow the standard cropping protocol, either by cropping the ROI and feeding it into CNN [16] or extracting CNN features using the full image and selecting those falling in the ROI [144].”