Neural network training method, target detection method, equipment, medium and product

The invention discloses a neural network training method, a target detection method, equipment, a medium and a product, and relates to the technical field of computers, and the method comprises the steps: receiving a verification picture set collected from a target detection scene, and generating a feature map of each picture in the verification picture set, the data type of each picture label comprises a detection frame and a plurality of key points, and calculating a plurality of prediction key points and key point scores of each feature map; and obtaining a key point regression result after determining that the feature map has a predicted key point according to the annotation data and the key point score, and adjusting structural parameters of the neural network according to an image detection result obtained according to the key point regression result and the annotation data. According to the method, whether the key points predicted by the neural network are reliable or not is judged by training the neur