Compressing the Input for CNNs with the First-Order Scattering Transform

@inproceedings{Oyallon2018CompressingTI,
  title={Compressing the Input for CNNs with the First-Order Scattering Transform},
  author={Edouard Oyallon and Eugene Belilovsky and Sergey Zagoruyko and Michal Valko},
  booktitle={ECCV},
  year={2018}
}
We study the first-order scattering transform as a candidate for reducing the signal processed by a convolutional neural network (CNN). We show theoretical and empirical evidence that in the case of natural images and sufficiently small translation invariance, this transform preserves most of the signal information needed for classification while substantially reducing the spatial resolution and total signal size. We demonstrate that cascading a CNN with this representation performs on par with… 
View PDF on arXiv
19 Citations
Background Citations
13
Methods Citations
6

19 Citations

Harmonic Networks for Image Classification
TLDR
This paper extensively validate the approach and shows that the introduction of harmonic blocks into state-of-the-art CNN models results in improved classification performance on CIFAR and ImageNet datasets.
Harmonic Convolutional Networks based on Discrete Cosine Transform
Boosting Deep Neural Networks with Geometrical Prior Knowledge: A Survey
TLDR
This survey tries to give a concise overview about different approaches to incorporate geometrical prior knowledge into DNNs, and tries to connect those methods to the field of 3D object detection for autonomous driving, where they expect promising results applying those methods.
The Unreasonable Effectiveness of Patches in Deep Convolutional Kernels Methods
TLDR
This work highlights the importance of a data-dependent feature extraction step that is key to the obtain good performance in convolutional kernel methods, and extensively study its effect, demonstrating it is the key ingredient for high performance of these methods.
Harmonic Networks: Integrating Spectral Information into CNNs
TLDR
This paper proposes harmonic blocks that produce features by learning optimal combinations of spectral filters defined by the Discrete Cosine Transform that are used to replace conventional convolutional layers to construct partial or fully harmonic CNNs.
Decoupled Greedy Learning of CNNs
TLDR
Decoupled Greedy Learning is considered, based on a greedy relaxation of the joint training objective, recently shown to be effective in the context of Convolutional Neural Networks (CNNs) on large-scale image classification, and it is shown that it can lead to better generalization than sequential greedy optimization.
IN DEEP CONVOLUTIONAL KERNELS METHODS
TLDR
This work highlights the importance of a data-dependent feature extraction step that is key to the obtain good performance in convolutional kernel methods, and extensively study its effect, demonstrating it is the key ingredient for high performance of these methods.
Classification with Joint Time-Frequency Scattering.
TLDR
The joint time-frequency scattering transform is introduced, a locally time-shift invariant representation which characterizes the multiscale energy distribution of a signal in time and frequency which may be implemented as a deep convolutional neural network whose filters are not learned but calculated from wavelets.
Joint Time–Frequency Scattering
TLDR
The joint time–frequency scattering transform is introduced, a time-shift invariant representation that characterizes the multiscale energy distribution of a signal in time and frequency that may be implemented as a deep convolutional neural network whose filters are not learned but calculated from wavelets.
Interferometric Graph Transform: a Deep Unsupervised Graph Representation
TLDR
The IGT is proposed, which is a new class of deep unsupervised graph convolutional neural network for building graph representations, and it is shown that the learned representation consists of both discriminative and invariant features, thanks to a novel greedy concave objective.
...
...

References

SHOWING 1-10 OF 41 REFERENCES
Very Deep Convolutional Networks for Large-Scale Image Recognition
TLDR
This work investigates the effect of the convolutional network depth on its accuracy in the large-scale image recognition setting using an architecture with very small convolution filters, which shows that a significant improvement on the prior-art configurations can be achieved by pushing the depth to 16-19 weight layers.
Spectral Representations for Convolutional Neural Networks
TLDR
This work proposes spectral pooling, which performs dimensionality reduction by truncating the representation in the frequency domain, and demonstrates the effectiveness of complex-coefficient spectral parameterization of convolutional filters.
Faster Neural Networks Straight from JPEG
TLDR
A simple idea is proposed and explored: train CNNs directly on the blockwise discrete cosine transform (DCT) coefficients computed and available in the middle of the JPEG codec, modified to produce DCT coefficients directly, and evaluated on ImageNet.
Wavelet Pooling for Convolutional Neural Networks
Towards Image Understanding from Deep Compression without Decoding
TLDR
This study shows that accuracies comparable to networks that operate on compressed RGB images can be achieved while reducing the computational complexity up to $2\times, and finds that inference from compressed representations is particularly advantageous compared to inference from compression RGB images for aggressive compression rates.
ImageNet classification with deep convolutional neural networks
TLDR
A large, deep convolutional neural network was trained to classify the 1.2 million high-resolution images in the ImageNet LSVRC-2010 contest into the 1000 different classes and employed a recently developed regularization method called "dropout" that proved to be very effective.
Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks
TLDR
This work introduces a Region Proposal Network (RPN) that shares full-image convolutional features with the detection network, thus enabling nearly cost-free region proposals and further merge RPN and Fast R-CNN into a single network by sharing their convolutionAL features.
Mask R-CNN
TLDR
This work presents a conceptually simple, flexible, and general framework for object instance segmentation that outperforms all existing, single-model entries on every task, including the COCO 2016 challenge winners.
Densely Connected Convolutional Networks
TLDR
The Dense Convolutional Network (DenseNet), which connects each layer to every other layer in a feed-forward fashion, and has several compelling advantages: they alleviate the vanishing-gradient problem, strengthen feature propagation, encourage feature reuse, and substantially reduce the number of parameters.
Fisher vectors meet Neural Networks: A hybrid classification architecture
TLDR
A hybrid architecture that combines their strengths: the first unsupervised layers rely on the FV while the subsequent fully-connected supervised layers are trained with back-propagation, which significantly outperforms standard FV systems without incurring the high cost that comes with CNNs.
...
...