search

pUmpKin-Co / offical-IndexNet

This is the official implementation of the IndexNet.
Apache License 2.0
10 stars 1 forks source link

Issues of training both instance contrast and index contrast #4

Closed tulilin closed 1 year ago

tulilin commented 1 year ago

Hello, I appreciate your work and I have read your code. For self-supervised pretraining, either instance contrast or index contrast is trained in the code in my understanding. Is there code implementation of training both instance contrast and index contrast? And for implementation, are MLPs in instance contrast and ConvMLPs in index contrast weight-sharing? I am looking forward to your reply. Thank you!

pUmpKin-Co commented 1 year ago

Hi~I'm feel so sorry for my mistake. I miss the glboal branch (instance contrast) in the first released commit. I have added the possible implementation of training both instance contrast and index contrast in IndexNetModel/IndexNetContrastMo.py and trainer.py.

Moreover, the MLPs in instance contrast and ConvMLPs in index contrast are not weight-sharing (see IndexNetModel/IndexNetContrastMo.py).

Thank you for your interest to our work, more questions on the implementation are welcome !