Embodiments of this disclosure provide an object recognition method and apparatus and a single step object recognition neural network, in which by replacing a part of convolutional layers in a conventional single shot multibox detector (SSD) network with dilate convolution, limitation on a size of an input image may be reduced; by increasing connection paths, accuracy of object recognition may be improved; and by connecting the convolutional layers by using residual structures, a convergence effect may be ensured, and accuracy of object recognition may be improved.
展开▼