Scene understanding method based on multi-task learning

Abstract

The invention provides a scene understanding method based on multi-task learning. The method includes: multi-task learning having unstable homoscedasticity, multi-task likelihood function, and scene understanding model. The method includes the following steps: first executing weighted linear sum on the loss of each individual task, learning the optimal task weight, inducting a multi-task loss function, defining a probability model, defining the probability as the gaussian function of the average values that are output by the model, eventually modeling pixel-level learning regression and classified output, comprising semantic division, incidence division and depth regression. According to the invention, the scene understanding model can learn multi-task weight, is more advantageous than models that independently train each task in that the method herein reduces computing amount, increases learning efficiency and prediction precision and can be real-timely operated.

Description

technical field [0001] The invention relates to the field of scene understanding, in particular to a scene understanding method based on multi-task learning. Background technique [0002] Scene understanding can effectively help and improve the computer's ability to analyze and recognize complex and changeable indoor and outdoor scenes, and is one of the research hotspots in the field of computer vision. Scene understanding can usually be divided into two categories: local scene understanding and global scene understanding. The former focuses on the analysis and description of the distribution and category of the local area of ​​the scene, such as the identification and positioning of various types of local targets in the scene; the latter focuses on understanding the global attributes of the scene, such as scene classification. Both can deepen the computer's cognition and grasp of unknown scenes from different cognitive levels, and have broad application prospects in the f...

AI Technical Summary

Problems solved by technology

[0004] For the problem of unsatisfactory efficiency and prediction accuracy, the purpose of the present invention is to provide a scene understanding method based on multi-task learning, first perform the weighted linear sum of the losses of each individual task, learn the optimal task weight, and then derive A multi-task loss function that defines a probabilistic model that defines likelihood as a Gaussian function of the mean given by the model output, and finally builds models that learn regression and classification outputs at the pixel level, including semantic segmentation, instance segmentation, and deep regression

Images