This is the face recognizer to identify awesome Arado people. The work is done top of the wonderful FaceNet GitHub project.
- Download pre-trained model
- Create a models folder in the root of the repository and copy the trained model there
- Copy the images into the
./arado/raw/
folder, create a folder for each object/person you want the model to recognise and put the corresponding pictures there. - Run โalign_arado_data.shโ to align pictures and copy them into arado160
- Run โpython src/align/create_test_train_sets.pyโ to create test and train data.
- To train run:
train_arado.sh
- To test run:
test_arado.sh
- To recognize Arado people run:
recognize_arado.sh
These are the results we got out of our trained model.
0 anssi: 0.397 arado/arado_160_test/anssi/anssi.png
1 henrik: 0.404 arado/arado_160_test/henrik/arado-2.png
2 jarno: 0.565 arado/arado_160_test/jarno/pic1.png
3 jarno: 0.596 arado/arado_160_test/jarno/pic10.png
4 jarno: 0.293 arado/arado_160_test/jarno/pic11.png
5 jarno: 0.500 arado/arado_160_test/jarno/pic2.png
6 markus: 0.383 arado/arado_160_test/markus/markus 2.png
7 mika: 0.448 arado/arado_160_test/mika/arado-2.png
8 mikko: 0.545 arado/arado_160_test/mikko/arado-2.png
9 stefano: 0.361 arado/arado_160_test/stefano/arado-2.png
10 teppo: 0.765 arado/arado_160_test/teppo/teppo.png
11 timo: 0.422 arado/arado_160_test/timo/arado-2.png
12 ville: 0.263 arado/arado_160_test/ville/arado.png
Accuracy: 1.000