search

ptx9363 / BCNet

Code for "Low Shot Box Correction for Weakly Supervised Object Detection"
MIT License
12 stars 4 forks source link

About test #4

Closed wcy1122 closed 3 years ago

wcy1122 commented 4 years ago

Did you use multi-scale for testing to report your performance in IJCAI paper? And moreover, horizontal flip? Since I only found {400, 600} scales without horizontal in your code.

Thanks

ptx9363 commented 4 years ago

The code is exactly what we use in IJCAI paper. The multi-scale test setting is that same with "Weakly- and Semi-Supervised Object Detection with Expectation-Maximization Algorithm". We use {400, 600} due to gpu memory limit.