GoogLeNet is a deep convolutional neural network architecture introduced by Szegedy et al. that employs Inception modules — parallel convolutional filter banks of varying sizes concatenated at each layer — enabling the model to capture multi-scale spatial features while maintaining computational efficiency through dimensionality reduction. In crowd counting and density estimation, its deep representational capacity allows it to extract rich hierarchical features from complex scenes, making it a strong backbone for regression and classification tasks. In WiFi CSI-based human sensing, GoogLeNet serves as a benchmark deep learning model for evaluating activity recognition performance, with its Inception-based design offering a notable architectural contrast to simpler CNNs and residual networks used in the same comparisons.
Dictionary term