Facenet: what the steps to run the code?

[OneMorea]

don't know how to run the program,without "readme" l can't understand which to run first and which later.hope somebody can show me！！！

[ShownX]

pls check wiki page

[OneMorea]

???Excuse me,can u pls give me a hint or a website link? Still have no clue what to check @ShownX Thank u!

[anhhoangta]

FYI: https://github.com/davidsandberg/facenet/wiki

[OneMorea]

@anhhoangta Thank you so much .I still dont know how to run it by this "readme".Can anyone give me an in-detail explanation or a demo pls？？？Thank you!!!

[anhhoangta]

So, what do you want to do? Compare two or more images? Search a face in many faces?

[hanbim520]

Hi ,you should read wiki (https://github.com/davidsandberg/facenet/wiki )and need to read the code what you want to do.

[OneMorea]

Search a face in many faces!First,pre-train,then choose a photo for test, so l can get a photo as same as the input and recognize who it is. How should l do ???

[anhhoangta]

I think you have to understand the goal of the facenet. If you have own face database and you want to search a given face in that, you should make a classifier. But there is a distinction between training a facenet model for feature representation and training a model for classifying people with the facenet pre-trained model.
So, for a classifier, first you load the pre-trained model, put your own database as input of this model, the output are the representations, 128-dimensions, corresponding each images in database. After that, you can use some classifier algorithms: SVM, k-NN, small neural network,...
and save it for prediction later.
You should read the paper and source code for more details before apply it.

[OneMorea]

@anhhoangta Sorry,I cant undertstand it very well so far,but l really need to run it successfully right now.There are so many folders that I really dont know the specific steps.Can you pls tell me the detailed steps to run it successfully or give me a demo video?thx!!!

[anhhoangta]

@XhedvaH if you need it right now, I think you should use openface: http://cmusatyalab.github.io/openface/
You can make your own classifier easily rely on their demo, very clearly!

[harshdhaware]

Gone through whole read me and wiki still not able to run the code. Please give me steps how to run your code...
Thank you in advance.

[Musawar71]

I am having isue in download_and_extract

[Musawar71]

please guide me

[jdowdells]

Hey Musawar71. I get really frustrated with some of this stuff. This is the second thing I've looked it for AI on github and it's never obious what to do. A simple explanation would be useful but for some reason these very intelligent people like to make us mere mortals work hard for the information. Anyway, after a while I worked out you can do stuff like:

python src/compare.py data/20180402-114759/ ./data/images/Anthony_Hopkins_0001.jpg ./data/lfw/Aaron_Peirsol/Aaron_Peirsol_0002.jpg

this gives an output like:

Distance matrix
0 1
0 0.0000 1.4831
1 1.4831 0.0000

but if you compare:

python src/compare.py data/20180402-114759/ ./data/images/Anthony_Hopkins_0001.jpg ./data/images/Anthony_Hopkins_0002.jpg

you get:
Distance matrix
0 1
0 0.0000 0.8396
1 0.8396 0.0000

and

python src/compare.py data/20180402-114759/ ./data/images/Anthony_Hopkins_0001.jpg ./data/images/Anthony_Hopkins_0001.jpg

you get

Distance matrix
0 1
0 0.0000 0.0000
1 0.0000 0.0000

So I'd conclude the the lower the numbers the more chance of a match.

If people want others to take up their work and try and integrate it into apps/products more comprehensive usage information is required.

We can't all magically guess what to do.

[jdowdells]

P.S I should have said, 1st thing I did was to run:

python test/train_test.py

in the facenet directory.

Also, when I set up the environment I had to use bazel 10.0