PyTorch implementation for Module Collection of Image-Text Retrieval for Further Exploration.
Importantly, the code (completed in September 2022) is not comprehensive and is not executable directly.
It functions as a compilation of popular modules, designed to ease adaptation for other domains.
We welcome any improvements and supplements by pulling requests to enhance the functionality of this code. Feel free to promote and share your papers during this collaborative process.
Basic Aggregation, Sequential GRU, Global Attention, Generalized Pooling, etc.
Like-Cosine Attention, Focal Attention, Relation-wise Attention,
Recurrent Attention, Transformer Attention, Bilinear Attention, etc.
Scalar Representation: Inner-product Similarity, Order-embedding Similarity, etc.
Vector Representation: Block Similarity, Symmetric or Asymmetric Similarity, etc.
Graph-based Aggregation: Local Alignments Enhancement, Global Alignments Guidance, etc.
Attention-based Aggregation: Local Alignments Filtration, Guidance Alignments Aggregation, etc.
Birank Loss, CMPL Loss, Binary Cross-entropy Loss, Angular Loss, etc.
If this code is useful for your research, please cite the relative papers in Awesome_Matching_Pretraining_Transfering.