Data Annotation Method and Apparatus, and Device

An apparatus comprising:

• a memory configured to store instructions;
• and
• one or more processors coupled to the memory and configured to execute the instructions to cause the apparatus to:
• receive, from a user, first to-be-annotated data, wherein the first to-be-annotated data comprises a first image, a first point cloud, and a first depth map;
• obtain, from a multimodal foundation model library, a target vertical-domain multimodal foundation model corresponding to a first service scenario associated with the first to-be-annotated data, wherein the apparatus is used in a cloud platform deployed with the multimodal foundation model library and a downstream task model library.
• receive, from the user, a first downstream task selection message indicating a first annotation type performed on the first to-be-annotated data;
• obtain, from the downstream task model library, a target downstream task model that corresponds to the first service scenario and that is for performing the first annotation type;
• use the target vertical-domain multimodal foundation model to extract a feature from the first to-be-annotated data;
• and
• use the target downstream task model to output based on the feature annotated data corresponding to the first to-be-annotated data.

A computer program product comprising computer-executable instructions that are stored on a non-transitory computer-readable storage medium and that when executed by one or more processors, cause an apparatus to:

• receive, from a user, first to-be-annotated data wherein the first to-be-annotated data comprises a first image, a first point cloud, and a first depth map;
• obtain, from a multimodal foundation model library, a target vertical-domain multimodal foundation model corresponding to a first service scenario associated with the first to-be-annotated data.
• receive, from the user, a first downstream task selection message indicating a first annotation type performed on the first to-be-annotated data;
• obtain, from a downstream task model library, a target downstream task model that corresponds to the first service scenario and that performs the first annotation type;
• use the target vertical-domain multimodal foundation model to extract a feature from the first to-be-annotated data;
• and
• use the target downstream task model to output based on the feature annotated data corresponding to the first to-be-annotated data.