Gini exponent-based domain adaptive semantic segmentation method

Abstract

The invention discloses a Gini exponent-based domain adaptive semantic segmentation method, which comprises the following steps of: measuring uncertainty of output prediction by utilizing a Gini exponent, performing uncertainty measurement and constraint on output prediction of a target domain on an output layer, reducing the difference between a source domain and the target domain in category distribution, and performing inter-domain adaptation; dividing a target domain sample set into two subsets according to an uncertainty measurement result of the Gini index for target domain prediction, training an intra-domain adaptive segmentation network for a sample corresponding to intra-domain high-confidence prediction by using a pseudo tag as weak supervision information, and calculating a Gini index graph for output prediction of the two subsets of the intra-domain adaptive segmentation network; a Gini exponent graph calculated by a low-confidence sample is constrained, a discriminator Dtis used for discriminating which subset the Gini index graph belongs to, the difference in a target domain is reduced based on an adversarial thought, and the semantic annotation precision is improved. Compared with the prior art, the semantic annotation accuracy of the target domain is remarkably improved.

Description

technical field [0001] The invention relates to an unsupervised field adaptive semantic tagging method, in particular to a field adaptive semantic segmentation method based on the Gini index, which belongs to the field of pattern recognition and computer vision and can be applied to automatic driving and robot visual navigation technologies. Background technique [0002] Unsupervised domain adaptive semantic segmentation uses labeled source domain data and unlabeled target domain data for training, and learns a model that has a better semantic annotation effect on target domain images. Accurate unsupervised domain-adaptive semantic segmentation is crucial for many applications, such as autonomous driving, robot navigation, etc. [0003] The main problem to be solved by unsupervised domain adaptation is how to reduce the difference between the source domain and the target domain. The usual strategies include: inter-domain adaptation of the input space, inter-domain adaptation...

AI Technical Summary

Problems solved by technology

Pseudo-label thresholds are usually artificially set empirically, requiring extensive experimentation and potentially introducing noise

Images