Skip to content

Latest commit

 

History

History
20 lines (12 loc) · 1013 Bytes

README.md

File metadata and controls

20 lines (12 loc) · 1013 Bytes

Based on Neuraltalk2 and AdaptiveAttention. Special thanks to the authors!

Testing codes and related codes will be added gradually.

When training using contrastive learning, better pre-train both the target and the reference.

Brief Explanation

  • adaptive attention: model structure for adaptive attention
  • neuraltalk2: model structure for neuraltalk2
  • misc: codes for DataLoader, util functions, and codes related to contrastive learning
    • cl_adaptiveattention.lua: codes for using adaptiveattention as both target model and reference model
    • cl_cross.lua: codes for using adaptiveattention as reference model, and using neuraltalk2 as target model
  • cl_train_adaptiveattention.lua: codes for training adaptiveattention
  • cl_train_cross.lua: codes for training neuraltalk2, using adaptiveattention as reference model