Model training method and device and drivable area detection method and device

CN121789185AActive Publication Date: 2026-04-03CHENGDU TIANFU INVO TECHNOLOGY CO LTD
View PDF 8 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-03-04
Publication Date
2026-04-03

AI Technical Summary

Technical Problem

In existing technologies, the collaborative optimization effect of multiple loss functions is highly dependent on manually setting weights, resulting in poor generalization ability.

Method used

A weight allocation structure is adopted to automatically generate loss function weights based on feature maps. By sharing the feature maps of the backbone network, the parameters of the backbone network and the region segmentation structure are optimized, and the weights are dynamically adjusted to adapt to the loss optimization needs of different scenarios.

Benefits of technology

It reduces the cost of manual debugging, improves the model's generalization ability, and ensures segmentation accuracy and robustness in different scenarios.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN121789185A_ABST
    Figure CN121789185A_ABST
Patent Text Reader

Abstract

The invention provides a model training method and device and a drivable area detection method and device, and relates to the technical field of image recognition, and the method comprises the steps: inputting an image sample into a backbone network, and obtaining a feature map outputted by the backbone network; inputting the feature map into a region segmentation structure to obtain a segmentation map output by the region segmentation structure; inputting the feature map into a weight distribution structure to obtain weights of a plurality of loss functions output by the weight distribution structure; calculating a total loss value based on each loss function, the weight of the loss function and the segmentation map; updating parameters of the backbone network and parameters of the region segmentation structure based on the total loss value; and locking the parameters of the backbone network and the parameters of the region segmentation structure, and updating the parameters of the weight distribution structure based on the total loss value. The generalization ability of the model can be improved.
Need to check novelty before this filing date? Find Prior Art

Citation Information

Patent Citations

  • Training method of multi-task learning deep network and target detection method and device

    CN115359334A

  • Three-dimensional target identification method based on multi-view target identification network

    CN119516258A

  • Retinal blood vessel image segmentation method based on dynamic weight distribution

    CN120260087A

  • Medical image segmentation method based on adaptive anisotropic convolution

    CN120726076A

  • Simulation-to-reality calibration method based on multi-modal loss and dynamic weighting

    CN120805744A