Collaboratively Weighting Deep and Classic Representation via L2 Regularization for Image Classification

摘要: Deep convolutional neural networks provide a powerful feature learning capability for image classification. The deep features can be utilized to deal with many understanding tasks like classification and object recognition. However, the robustness obtained in one dataset hardly reproduced other domain, which leads inefficient models far from state-of-the-art. We propose collaborative weight-based (DeepCWC) method resolve this problem, by providing novel option fully take advantage of classic machine learning. It firstly performs L2-norm based representation on original images, as well extracted CNN models. Then, two distance vectors, pair linear representations, are fused together via weight. This weight enables representations weigh each other. observed complementarity between series experiments 10 facial datasets. proposed DeepCWC produces very promising results, outperforms benchmark methods, especially ones claimed Fashion-MNIST. code is going published our public repository.