sephirothhua/Lite_Upsample_Network_Carafe-unoffical-

The Lite Upsample Network Carafe unoffical

• This is an (unofficial) implementation of CARAFE: Content-Aware ReAssembly of FEatures with pytorch.
• The network is a light upsample module which performs well and has less parameters.

Questions

• The implementation of Context aware Reassembly Module maybe not the best. Is there any better ways to generate stride features?

Done

• The module has been tested on my own yolo and imporved about 1mAP on VOC 2007 test dataset. The code will be relesed later.

TODO

• Check if bn and relu in Context aware Ressembly can imporve the module.