Face Recognition

Qualeams

Mayelana nalolu hlelo lokusebenza

Ubuso Recognition ingasetshenziswa kanye nohlaka yokuhlola izindlela ukuqashelwa ubuso eziningana kuhlanganise Neural Networks nge TensorFlow futhi Caffe.

Kuhlanganisa elandelayo preprocessing algorithm:
- ufiphazo
- Nqampuna
- Eye Ukuhambisana
- Gamma Ukulungiswa
- Umehluko Gaussians
- esiyiqili Elihehayo
- Iphethini Binary Local
- Histogramm zokulinganiswa (ingasetshenziswa kuphela uma ukwenza mhlophe isetshenziswa kakhulu)
- Shintsha usayizi

Ungakhetha isici isizinda kanye ngezigaba izindlela ezilandelayo:
- Eigenfaces nge Umakhelwane eliseduze
- Image Reshaping nge Ukusekela Vector Machine
- TensorFlow nge SVM noma KNN
- Caffe nge SVM noma KNN

Leli bhukwana ingatholakala lapha https://github.com/Qualeams/Android-Face-Recognition-with-Deep-Learning/blob/master/USER%20MANUAL.md

Okwamanje kuphela kumadivayisi armeabi-v7a futhi phezulu zisekelwa.

Ukuze isipiliyoni best in ukuqashelwa imodi uphendukisa idivaysi kwesokunxele.
_______________________________________________________________

TensorFlow:

Uma ufuna ukusebenzisa imodeli Tensorflow Inception5h, ulilande lapha:
https://storage.googleapis.com/download.tensorflow.org/models/inception5h.zip

Khona-ke ukukopisha ifayela "tensorflow_inception_graph.pb" ukuze "/ sdcard / Isithombe / facerecognition / idatha / TensorFlow"

Sebenzisa lezi izilungiselelo ezizenzakalelayo isiqalo:
Inombolo amakilasi: 1001 (ayifanele njengoba asisebenzisi ungqimba wokugcina)
Wokufaka Usayizi: 224
Image kusho: 128
usayizi lokukhipha: 1024
Wokufaka ungqimba: okokufaka
Lokukhipha ungqimba: avgpool0
ifayela Model: tensorflow_inception_graph.pb
-------------------------------------------------- -------------------------------------------------- -----
Uma ufuna ukusebenzisa imodeli VGG Face yokuchaza, ulilande lapha:
https://www.dropbox.com/s/51wi2la5e034wfv/vgg_faces.pb?dl=0

Isexwayiso: Le modeli lusebenza kuphela kumadivayisi okungenani 3 GB noma RAM.

Khona-ke ukukopisha ifayela "vgg_faces.pb" ukuze "/ sdcard / Isithombe / facerecognition / idatha / TensorFlow"

Sebenzisa lezi izilungiselelo ezizenzakalelayo isiqalo:
Inombolo amakilasi: 1000 (ayifanele njengoba asisebenzisi ungqimba wokugcina)
Wokufaka Usayizi: 224
Image kusho: 128
usayizi lokukhipha: 4096
ungqimba wokufaka: Isimeli
Lokukhipha ungqimba: fc7 / fc7
ifayela Model: vgg_faces.pb
_______________________________________________________________

Caffe:

Uma ufuna ukusebenzisa imodeli VGG Face yokuchaza, ulilande lapha:
http://www.robots.ox.ac.uk/~vgg/software/vgg_face/src/vgg_face_caffe.tar.gz

Isexwayiso: Le modeli lusebenza kuphela kumadivayisi okungenani 3 GB noma RAM.

Khona-ke ukopishe amafayela "VGG_FACE_deploy.prototxt" futhi "VGG_FACE.caffemodel" ukuze "/ sdcard / Isithombe / facerecognition / idatha / Caffe"

Sebenzisa lezi izilungiselelo ezizenzakalelayo isiqalo:
Amanani Mean: 104, 117, 123
Lokukhipha ungqimba: fc7
ifayela Model: VGG_FACE_deploy.prototxt
Izisindo efayelini: VGG_FACE.caffemodel

_______________________________________________________________

Amafayela ilayisensi ingatholakala lapha https://github.com/Qualeams/Android-Face-Recognition-with-Deep-Learning/blob/master/LICENSE.txt futhi lapha https://github.com/Qualeams/Android- ubuso-Recognition-ne-Deep-Learning / blob / master / NOTICE.txt

Yini entsha

- Switch from building Tensorflow from source to using the Jcenter library
- Included optimized_facenet model and changed default settings to use TensorFlow by default