dgs.models.embedding_generator

Modules for generating embeddings for Re-Identification (‘ReID-Embeddings’).

May either be for visual Re-ID or for pose-based Re-ID. Given some input data, predict something like an embedding vector to describe the input. This vector is then used to compare it to other embedding vectors using a similarity function. The vector of one specified target should be close to other generated vectors of this target, but as far as possible from independent targets.

For using an EmbeddingGeneratorModule as similarity, see the description of SimilarityModule.

Module Functions

get_embedding_generator(name)	Given the name or a new instance of an embedding generator module, return the type.
register_embedding_generator(name, new_eg)	Register a new embedding generator in EMBEDDING_GENERATORS, to be able to use it from configuration files.

Modules

embedding_generator	Base module for models used for embedding generation.
pose_based	Different pose based embedding generators.
torchreid	Use a model out of the torchreid package as an embedding generator.