Transformation Pursuit for Image Classification

By Mattis Paulin, Jérôme Revaud, Zaïd Harchaoui, Florent Perronnin, Cordelia Schmid. 2014

107 citations recorded by the source corpus.

Subfields: Artificial Intelligence, 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 Artificial Intelligence

In Computer Vision and Pattern Recognition

Citations in the corpus

Recent advances in convolutional neural networks (2015, background): “[113] propose a greedy strategy that selects the”
APAC: Augmented PAttern Classification with Neural Networks (2015, methodology): “Paulin et al . proposed a novel method for creating augmented datasets [15].”
Return of the Devil in the Details: Delving Deep into Convolutional Nets (2014, background): “These improvements could also transfer to shallow representations such as the IFV, potentially explaining a part of the performance gap [22].”
Fisher vectors meet Neural Networks: A hybrid classification architecture (2015, methodology): “To this end, we apply ITP [35] to select 4 transformations (on top of the original images), and apply them at training and test time.”
MoCap-guided Data Augmentation for 3D Pose Estimation in the Wild (2016, background): “Such data augmentation has proven to be a crucial stage, especially for training deep architectures.”
Towards Reversal-Invariant Image Representation (2017, methodology): “By data augmentation we mean to generate a reversed copy for each training/testing image, use the augmented set to train the model, test with both original and reversed samples, and predict the label with a soft-max function (Paulin et al. 2014).”
Incorporating visual adjectives for image classification (2016, background): “…are extracted, it is possible to fuse them at the descriptor-stage (mixing both descriptors on the image plane) [29], at the feature-stage (concatenating encoded feature vectors into one) [26,30], or at the classi ﬁ er-stage (fusing individual classi ﬁ cation scores for ﬁ nal judgement) [31].”
Billion-scale semi-supervised learning for image classification (2019, background): “Several recent strategies have been successful at improving classification accuracy by learning the transformation learned for augmentation [7, 31] or by synthesizing mixtures of images [45].”
RIFD-CNN: Rotation-Invariant and Fisher Discriminative Convolutional Neural Networks for Object Detection (2016, background): “Consequently, our work is also partly related with the ideas of learning invariant features such as [40-46] and the approaches incorporating discriminative terms into model training such as [47, 48].”
Sympathy for the Details: Dense Trajectories and Hybrid Classification Architectures for Action Recognition (2016, unknown): “Common data augmentation techniques for images include the use of random horizontal flipping [52,26], random cropping [52], and even automatically determined transformations [54].”
Face recognition using modified deep learning neural network (2017, methodology): “Data augmentation is one other easy method which has been used to tackle the problem of dearth of training data [11], [12].”
Towards textual data augmentation for neural networks: synonyms and maximum loss (2019, background): “Most of the work has been done with visual data [5, 6, 13, 32] yet less with textual data, although more and more papers have been published on deep learning for NLP [7, 41, 52] (see Figure 1).”