Low-altitude visual sample generation model training method, low-altitude visual sample generation method, electronic device and storage medium

Abstract

Embodiments of the present application provide a low-altitude visual sample generation model training method, a low-altitude visual sample generation method, an electronic device and a storage medium, relating to the technical field of image processing. Freeze the discriminator, calculate the generator loss and update the parameters of the generator based on at least one target sample category, freeze the generator, obtain a batch of real samples and the false samples generated by the generator, assign pseudo-labels to the false samples which are independent of the label range of the real samples, input the real samples and the false samples into the discriminator for feature extraction, obtain real sample features and false sample features, respectively construct positive and negative sample pairs, calculate real sample contrast loss and false sample contrast loss, update the parameters of the discriminator, alternately iterate the steps of updating the parameters of the generator and updating the parameters of the discriminator until convergence, and output the trained generator as a low-altitude visual sample generation model. Thus, the inter-class discriminability and intra-class consistency of the generated samples in the feature space can be improved.

Images