search

DirtyHarryLYL / Transferable-Interactiveness-Network

Code for Transferable Interactiveness Knowledge for Human-Object Interaction Detection. (CVPR'19, TPAMI'21)
MIT License
227 stars 41 forks source link

The shape of pool5_O in early and late fusion #54

Closed YSN1011 closed 4 years ago

YSN1011 commented 4 years ago

Hello, and thank you for your job well done. I have a problem with the code in this article.In late fusion, the pool5_O shape is (46,7,7,1024).In early fusion, pool5_O was in the shape of (16,7,7,1024).I have the following questions:

  1. Pool5_O is the same production code in both merges, so am I missing something?Why are the sample sizes inconsistent?
  2. Why are the sample sizes selected differently in the two different fusion methods? I hope we can get a solution. Thank you very much.
Foruck commented 4 years ago

Could you please specify what you referred to as 'early fusion' and 'late fusion'? And the mentioned size infonsistency might be the result of Line 405 in lib/networks/TIN_HICO.py, where only the first self.H_num object bboxes are used for ROI pooling.