Object positioning method and device, equipment and medium
The embodiment of the invention provides an object positioning method and device, equipment and a medium, and the method comprises the steps: integrating multi-level text representation and image representation from a level angle in a forward process, and achieving the multi-mode self-adaption; in t...
Saved in:
Main Authors | , , , , , , |
---|---|
Format | Patent |
Language | Chinese English |
Published |
03.05.2024
|
Subjects | |
Online Access | Get full text |
Cover
Loading…
Summary: | The embodiment of the invention provides an object positioning method and device, equipment and a medium, and the method comprises the steps: integrating multi-level text representation and image representation from a level angle in a forward process, and achieving the multi-mode self-adaption; in the reverse process, under the condition that the weight matrix of the deep network layer group of the image encoder is frozen, the low-rank matrix of the shallow network layer group is firstly updated, the weight matrix of the shallow network layer group is frozen, the network layer groups are gradually increased, and the process of updating the low-rank matrix is repeated after the network layer groups are increased every time; through hierarchical decoupling, the learning rate of the image encoder is changed in different adaptation stages, the image encoder is ensured to gradually adapt to deep features from shallow features, and interaction and alignment of fine-grained cross-modal features are realized. The dif |
---|---|
Bibliography: | Application Number: CN202410382411 |