search

FlyingRoastDuck / MetaCam_DSCE

Code for our CVPR 2021 paper "MetaCam+DSCE"
59 stars 12 forks source link

About the use of outliers #5

Open haoni0812 opened 1 year ago

haoni0812 commented 1 year ago

Hi, thanks for sharing your work. I noticed that the outliers seem not to be used in the code (as shown below). However, according to the description in the paper, outliers should be assigned to the nearest pseudo category. I would like to know where this operation is carried out.

in em.py: nums.indexadd(0, labels[labels != -1], torch.ones(labels[labels != -1].shape[0], 1).float().cuda())

FlyingRoastDuck commented 10 months ago

Hi, Sorry for the late reply. You can find the ``outlier merging" operation here:

https://github.com/FlyingRoastDuck/MetaCam_DSCE/blob/master/reid/utils/__init__.py#L43

if you have further questions, please contact me through email. I cannot reply to you timely with github issues.