TY - GEN
T1 - Introduction of explicit visual saliency in training of deep CNNs
T2 - 16th International Conference on Content-Based Multimedia Indexing, CBMI 2018
AU - Obeso, Abraham Montoya
AU - Benois-Pineau, Jenny
AU - Garcia Vazquez, Mireya Sarai
AU - Ramirez Acosta, Alejandro A.
N1 - Publisher Copyright:
© 2018 IEEE.
PY - 2018/10/30
Y1 - 2018/10/30
N2 - Introduction of visual saliency or interestingness in the content selection for image classification tasks is an intensively researched topic. It has been namely fulfilled for feature selection in feature-based methods. Nowadays, in the winner classifiers of visual content such as Deep Convolutional Neural Networks, visual saliency maps have not been introduced explicitly. Pooling features in CNNs is known as a good strategy to reduce data dimensionality, computational complexity and summarize representative features for subsequent layers. In this paper we introduce visual saliency in network pooling layers to spatially filter relevant features for deeper layers. Our experiments are conducted in a specific task to identify Mexican architectural styles. The results are promising: proposed approach reduces model loss and training time keeping the same accuracy as the base-line CNN.
AB - Introduction of visual saliency or interestingness in the content selection for image classification tasks is an intensively researched topic. It has been namely fulfilled for feature selection in feature-based methods. Nowadays, in the winner classifiers of visual content such as Deep Convolutional Neural Networks, visual saliency maps have not been introduced explicitly. Pooling features in CNNs is known as a good strategy to reduce data dimensionality, computational complexity and summarize representative features for subsequent layers. In this paper we introduce visual saliency in network pooling layers to spatially filter relevant features for deeper layers. Our experiments are conducted in a specific task to identify Mexican architectural styles. The results are promising: proposed approach reduces model loss and training time keeping the same accuracy as the base-line CNN.
KW - CNNs
KW - Pooling
KW - Saliency Maps
UR - http://www.scopus.com/inward/record.url?scp=85057031191&partnerID=8YFLogxK
U2 - 10.1109/CBMI.2018.8516465
DO - 10.1109/CBMI.2018.8516465
M3 - Contribución a la conferencia
AN - SCOPUS:85057031191
T3 - Proceedings - International Workshop on Content-Based Multimedia Indexing
BT - 16th International Conference on Content-Based Multimedia Indexing, CBMI 2018
PB - IEEE Computer Society
Y2 - 4 September 2018 through 6 September 2018
ER -