Deep Roto-Translation Scattering for Object Classification

Abstract

Dictionary learning algorithms or supervised deep convolution networks have considerably improved the efficiency of predefined feature representations such as SIFT. We introduce a deep scattering convolution network, with predefined wavelet filters over spatial and angular variables. This representation brings an important improvement to results previously obtained with predefined features over object image databases such as Caltech and CIFAR. The resulting accuracy is comparable to results obtained with unsupervised deep learning and dictionary based representations. This shows that refining image representations by using geometric priors is a promising direction to improve image classification and its understanding.

9 pages, 3 figures, CVPR 2015 paper

References in corpus (6)

• Intriguing properties of neural networks
• Spatial Pyramid Pooling in Deep Convolutional Networks for Visual Recognition
• Deeply-Supervised Nets
• Rich feature hierarchies for accurate object detection and semantic segmentation
• Building high-level features using large scale unsupervised learning
• Group Invariant Scattering

Cited by in corpus (3)

• Understanding Adversarial Training: Increasing Local Stability of Neural Nets through Robust Optimization
• Discrete Deep Feature Extraction: A Theory and New Architectures
• Deep Haar Scattering Networks