DpicNet is a deep convolutional neural network for image classification trained towards Intel Multi-class Image Classification Dataset. DpicNet comprises of convolutional part where features are extracted from images and classification part where features are passed into fully-connected or dense network to classify. As DpicNet follows transfer learining procedure, the convolutional part is transferred from existing deep neural network model trained on large-scale image classification such as ImageNet dataset. Xception is chosen and its convolutional is transferred and freezed, so that the weights of the convolutional part is not trained. The top part, fully-connected or dense network, is newly constructed for the purpose of training towards Intel Multi-class Image Classification Dataset. Many settings for the fully-connected network were experimented, but the best setting can achieve approcimately 98% accuracy on training dataset and 86% accuracy on testing dataset. The best setting is served using TensorFlow Serving along with Flask. More details are available on Github by following the project link below.