Do we need to normalize the image before feeding into the re-id resnet?

phil-bergmann / tracking_wo_bnw

Implementation of "Tracking without bells and whistles” and the multi-object tracking "Tracktor"

https://arxiv.org/abs/1903.05625

GNU General Public License v3.0

818 stars 196 forks source link

Do we need to normalize the image before feeding into the re-id resnet? #103

Closed alanzty closed 3 years ago

alanzty commented 4 years ago

I found a yaml file coming with pretrained weights of re-id network. It has stated commonly used normalization normalize_mean:

0.485
0.456
0.406 normalize_std:
0.229
0.224
0.225 However, in the test time, I found that the image is not normalized. Since I am about to use this as a feature extractor, I want to be careful here. Do we actually need to normalize the image?

timmeinhardt commented 3 years ago

The images are normalized and test time in the corresponding dataset code. So yes, you have to normalize them as well if you want to use our reID network as a feature extractor.

timmeinhardt commented 3 years ago

Can we close this issue?