I-DeepDetect yisoftware yamahhala ithuthukiswe nguJoliBrain, inhloso yawo ukwenza izinto ezintsha zakamuva zifinyeleleke futhi zisetshenziswe ekufundeni okujulile (ukufunda okujulile) kuyatholakala futhi kuvumela ukuhlanganiswa kwezicelo.
Ukuthola okujulile inezinhlelo ezimbili zamahhala: omunye wabo ngu iseva ebhalwe ku-C ++ 11 ene-REST API, evumela ukufinyelela kumitapo yolwazi engaphansi kweCaffe, Caffe2, Tensorflow, Dlib, NCNN, njll. Enye iyisiteji sewebhu ukuqeqesha, ukuhlela nokusebenzisa amamodeli wakho njengamazwibela amancane ekhodi.
Ukutholwa komcimbi okuzenzakalelayo kusuka kumasiginali uchungechunge lwesikhathi lunokusetshenziswa okubanzi. Izindlela zokuthola zendabukos ukuthola imicimbi ngokuyinhloko ngokusetshenziswa kokufana nokulungiswa kwedatha.
Lezo zindlela zingasebenzi kahle futhi zikhiqize ukucacisa okuphansi. Eminyakeni yakamuva, amasu okufunda ngomshini aguqule izizinda eziningi zesayensi nezobunjiniyela.
Ikakhulu, ukusebenza kokutholwa kwento kumininingwane yesithombe esingu-2-D kuthuthukiswe kakhulu ngenxa yamanethiwekhi ajulile e-neural.
Mayelana nepulatifomu ye-DeepDetect
Ukuthola okujulile isebenzisa ukwesekwa kokufunda okujulile okugadiwe futhi okungagadwe kwezithombe, umbhalo nenye idatha, ngokugxila kubulula nokusebenziseka kalula, ukuhlolwa nokuxhumeka kuzinhlelo zokusebenza ezikhona.
Ixhasa ukuhlukaniswa, ukutholwa kwento, ukwehlukaniswa, ukubuyela emuva, ama-autoencoders, nokunye okuningi.
Phakathi kwezici zayo eziyinhloko kukhona
- Izinga eliphakeme le-API lokufunda ngomshini nokufunda okujulile
- ukusekelwa kweCaffe, Tensorflow, XGBoost ne-T-SNE
- Ukuhlukaniswa, ukuhlehliswa, ama-autoencoders, ukutholwa kwento, ukuhlukaniswa.
- Ifomethi yokuxhumana ye-JSON
- umtapo wezincwadi weklayenti elikude le-python
- Iseva ezinikezele ngokusekelwa kwamakholi wokuqeqesha asynchronous.
- Ukusebenza okuphezulu, ukuzuza kuma-CPU amaningi kanye ne-GPU
- ukusesha kokufana okwakhelwe ngaphakathi ngokushumeka kwe-neural
- Isixhumi sokuphatha amafayela we-CSV ngamandla wokucubungula
- Isixhumi sokuphatha amafayela wombhalo, imisho namamodeli asuselwa kubalingiswa.
- Isixhumi sokuphatha ifomethi yefayela le-SVM yedatha eyingcosana
- ngaphandle kokuncika nokuvumelanisa kwe-database, lonke ulwazi nemingcele yemodeli ehlelekile futhi etholakala ohlelweni lwefayela
- Flexible template okukhipha ifomethi ukwenza lula ukuxhumana nezinhlelo zokusebenza zangaphandle
- Ukusekelwa kokubalwa okuncane nemisebenzi ku-GPU naku-CPU.
Mayelana ne-LiveDetect
I-LiveDetect iyi- ithuluzi elenzelwe ukucubungula kalula imifudlana yevidiyo yasendaweni enamamodeli wokufunda ajulile. Ikhodi ifunda izithombe ezibukhoma kusuka kwikhamera futhi icubungula ifreyimu ngayinye nge-DeepDetect.
Amacala okusetshenziswa komhlaba wangempela wamakhasimende e-DeepDetect ane-LiveDetect:
- Ukuphepha esizeni kanye nokuqashelwa kwesiza.
- Ukubhaliswa kwemoto kwe-OCR ezindaweni zokupaka.
- Ukutholwa kwamaphutha ezingxenyeni zokunemba ezenziwe.
Ungayifaka kanjani i-DeepDetect ku-Raspberry pi?
I-DeepDetect ingafakwa kuzingxenyekazi ezahlukahlukene (zombili kumaseva, njengamakhompyutha, ama-laptops ngisho naku-Raspberry Pi).
Kusuka kuwebhusayithi esemthethweni ye-DeepDetect singakwazi thola imiyalo yokufaka kwipulatifomu ngayinye esekelwayo.
Kulokhu, Sizofaka i-DeepDetect ku-Raspberry Pi yethu, nge-back-end ye-NCNN ne-LiveDetect, ithuluzi elisuselwa kuhlelo lwe-DeepDetect ecosystem lokucubungula ukulandelana kwamavidiyo. Lokhu kusivumela ukuthi sithole izinto ngesikhathi sangempela futhi sizibone ngeso lengqondo.
Amamodeli we-Deep Learning aqeqeshwe ngaphambilini ayatholakala kuma-desktop nezinhlelo ezishumekiwe njenge-Raspberry Pi.
Ukufaka i-DeepDetect server ku-Raspberry Pi, Sizosebenzisa iDocker ukuze kube lula nokusebenza okuhle.
Into yokuqala esizoyenza dala ifolda ye-DeepDetect Docker Container, lokhu sikwenza ngokuvula i-terminal bese siyisebenzisa:
mkdir $HOME/models
docker pull jolibrain/deepdetect_ncnn_pi3
docker run -d -p 8080:8080 -v $HOME/models:/opt/models jolibrain/deepdetect_ncnn_pi3
sudo apt-get install libjpeg-dev
Manje sizolanda i-LiveDetect bese siyifaka:
wget https://github.com/jolibrain/livedetect/releases/download/1.0.1/livedetect-rpi3
./livedetect-rpi3 \
--port 8080 \
--host 127.0.0.1 \
--mllib ncnn \
--width 300 --height 300 \
--detection \
--create --repository /opt/models/voc/ \
--init "https://www.deepdetect.com/models/init/ncnn/squeezenet_ssd_voc_ncnn_300x300.tar.gz" \
--confidence 0.3 \
-v INFO \
-P "0.0.0.0:8888" \
--service voc \
--nclasses 21
Ukunikezwa kuyatholakala ku-http: // localhost: 8888 phakathi kwamafreyimu amabili kuya kwamathathu ngesekhondi (FPS).
Uma ufuna ukusebenzisa i-LiveDetect kwikhompyutha yakho yedeskithophu, ungathola imiyalo kanye nemininingwane eminingi namasampula nge-LiveDetect etholakala ku-GitHub.
Isixhumanisi yilokhu.