These were successful in using parallel structures in the convolution network, which further increased the performance of models in the same competition. It was proposed and refined by Christian Szegedy, Vincent Vanhoucke, Sergey Ioffe, Jonathon Shlens, Zbigniew Wojna in the paper Rethinking the Inception Architecture for Computer Vision, https://arxiv.org/abs/1512.00567. The model structure for inception-v3 is as follows:
We can use this model in Keras:
from keras.applications.inception_v3 import InceptionV3def print_model(): """ Loads InceptionV3 model and prints model structure """ # create model model = InceptionV3(weights='imagenet') ...