Deep Convolutional Neural Network for Dog Breed classification is one of the best techniques to achieve this task.
Popular CNN architectures have been used namely Inception-ResNetV2, Resnet101, Resnet50 and InceptionV3 all pre-trained on the ImageNet dataset as robust feature extractors followed by the logistic regression classifier.
Across all the various models, InceptionResNetV2, Resnet101, Resnet50 and InceptionV3 were tested. The validation Multiclass Loss Score varied from 0.34708 (for Resnet101) to 5.11947 (for InceptionV3).