Facenet: why deal with image by "prewhiten" used in compare.py but not uesd for training

Created on 22 May 2017  路  2Comments  路  Source: davidsandberg/facenet

@Davidsandberg,
Hi, thank you for your wonderful work!
I have one question about prewhiten. hope you can give me some guide.

why prewhiten used to deal with image before inference in compare.py, but I have checked there is no prewhiten when train model, no matter in facenet_softmax.py or alin_dataset_mtcnn.py.
if so, the image format is not same with the image feeded in train, it doesn't matter?
if it doesn't matter, why used in compare.py.

this puzzled me for a long time, could you give me some explanation?

picture emily2he

Most helpful comment

Prewhiten is equivalent to tf.image.per_image_standardization(), which is used in training.

picture wangzzdeepir on 22 May 2017
😄2 👍2

All 2 comments

Prewhiten is equivalent to tf.image.per_image_standardization(), which is used in training.

wangzzdeepir picture wangzzdeepir on 22 May 2017
😄2 👍2

@wangzzdeepir thank you for your explanation!

emily2he picture emily2he on 22 May 2017
Was this page helpful?
0 / 5 - 0 ratings

Related issues

ouyangbei picture ouyangbei  路  4Comments
MaartenBloemen picture MaartenBloemen  路  3Comments
Leedonggeon picture Leedonggeon  路  3Comments
Zumbalamambo picture Zumbalamambo  路  3Comments
allahbaksh picture allahbaksh  路  3Comments