Method for training multi-target detection model, detection method and related device

The embodiment of the invention relates to the technical field of target detection, and discloses a method for training a multi-target detection model, a training set comprises a plurality of images with human faces, each image is labeled with a real label, and the real label comprises a hard label of at least one item. Respectively inputting the training set into at least one single-target detection model to obtain a soft label of each item corresponding to each image, and labeling each soft label corresponding to each image as a real label, so that the real label labeled by each image comprises at least one hard label and at least one soft label. And finally, inputting the training set into a convolutional neural network for training, and iteratively adjusting parameters to obtain the multi-target detection model. Through the training mode, classification and positioning of a plurality of items can be detected by using a plurality of target detection models, and the model deployment cost can be effectively