Confidence evaluation method, device, apparatus, medium and product

Abstract

The application provides a confidence evaluation method, device, equipment, medium and product, which can be applied to the technical fields of deep learning, image processing and computer vision. The method comprises the following steps: obtaining a target image and model weights of a target model; performing perturbation on the model weights according to a model weight perturbation strategy to obtain a plurality of perturbed models, and inputting the target image into the plurality of perturbed models respectively to determine a cognitive uncertainty degree related to the target model; performing perturbation on the target image according to an image perturbation strategy to obtain a plurality of perturbed images, and inputting the plurality of perturbed images into the target model respectively to determine a random uncertainty degree related to the target image; and determining a confidence image according to the cognitive uncertainty degree and the random uncertainty degree, wherein the confidence image comprises a plurality of pixel positions corresponding to the target image, and a pixel value of each pixel position is used to indicate a confidence degree of an image processing result obtained by processing the target image by using the target model.

Images