A Theoretical Analysis of Feature Pooling in Visual Recognition

By Y-Lan Boureau, Jean Ponce, Yann LeCun. 2010

1,404 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

Fixing the train-test resolution discrepancy (2019, neutral): “network to test resolutions signiﬁcantly higher than the training resolution. The authors show that this improves the network’s performance, in accordance with the conclusions drawn by Boureau et al. [6]. Similar pooling techniques have been employed in image retrieval for a few years [29, 38], wher”
Deformable part models are convolutional neural networks (2014, background): “Distance transform pooling, unlike the pooling functions in [2], is interesting because it has a learnable pooling region.”
Representation Learning: A Review and New Perspectives (2012, background): “Empirically the use of pooling seems to contribute significantly to improved classification accuracy in object classification tasks (LeCun et al., 1998b; Boureau et al., 2010, 2011).”
Deformable Convolutional Networks (2017, background): “Their capability of modeling geometric transformations mostly comes from the extensive data augmentation, the large model capacity, and some simple hand-crafted modules ( e.g ., max-pooling [1] for small translation-invariance).”
Beyond short snippets: Deep networks for video classification (2015, methodology): “We considered using both average pooling, and max-pooling which have several desirable properties as shown in [4].”
Text Understanding from Scratch (2015, methodology): “The analysis by (Boureau et al., 2010b) might shed some light on this.”
Robust Large Margin Deep Neural Networks (2016, neutral): “2708039 transformations such as translation [6], [7].”
Recent advances in convolutional neural networks (2015, background): “After several convolutional and pooling layers, there may be one or more fully-connected layers which aim to perform high level reasoning [8–10].”
Generalizing Pooling Functions in Convolutional Neural Networks: Mixed, Gated, and Tree (2015, background): “In the current deep learning literature, popular pooling functions include max, average, and stochastic pooling [3, 2, 40].”
Bag of visual words and fusion methods for action recognition: Comprehensive study and good practice (2014, neutral): “In [4], the authors presented a theoretical analysis of average pooling and max pooling.”
MultiGrain: a unified image embedding for classes and instances (2019, background): “Setting this exponent as p > 1 increases the contrast of the pooled feature map and focuses on the salient features of the image [2, 3, 7].”
CARER: Contextualized Affect Representations for Emotion Recognition (2018, methodology): “A 1-max pooling layer (Boureau et al., 2010) of size 3 is then applied to each feature map.”
XCiT: Cross-Covariance Image Transformers (2021, methodology): “All features have a constant stride of 8 or 16 based on the patch size, and the feature resolutions are adjusted to have strides of [4, 8, 16, 32], similar to ResNet-FPN backbones, where the downsampling is achieved by max pooling and the upsampling is obtained using a single transposed convolution ”
Rotation-invariant convolutional neural networks for galaxy morphology prediction (2015, positive): “A pooling layer reduces the dimensionality of a feature map by computing some aggregation function (typically the maximum or the mean) across small local regions of the input (Boureau et al. 2010), as shown in the right side of Figure 3.”
Modeling the spatial layout of images beyond spatial pyramids (2012, methodology): “This also applies to other image-level repre285 sentations and especially to the BOV as noted in [8, 9].”
Learning to Count Leaves in Rosette Plants (2015, methodology): “When all vectors zi are determined in an image, we use max-pooling to compute a global descriptor which reduces the size of the descriptor and also adds invariance to small local transformations [5, 20].”