Object detection method, neural network training method, apparatus and electronic equipment

摘要

The present application provides an object detection method, a neural network training method, an apparatus, and an electronic device. The object detection method predicts and obtains a plurality of fusion feature maps from an image to be processed by a deep convolutional neural network for target area box detection, wherein a plurality of fusion feature maps are obtained from a first subnet having at least one downsampling layer. And obtaining a plurality of second feature maps from a second subnet having at least one up-sampling layer, and fusing with each of the plurality of first feature maps and the plurality of second feature maps. Obtaining a feature map, and then further obtaining target area box data based on the plurality of fusion feature maps. These fused feature maps are based on these fused feature maps in order to effectively characterize semantic features of the upper layer (eg, layout, foreground information) and feature points of the lower layer (eg, small object information) in the image. It is possible to effectively extract target area box data of large and small objects included in an image, thereby improving the accuracy and robustness of object detection. [Selection diagram] Fig. 1