M'ndandanda wazopezekamo[Bisani][Show]
- 1. CelebFaces Attributes Dataset
- 2. DOTA
- 3. Google Facial Expression comparison dataset
- 4. Genome Yowoneka
- 5. LibriSpeech
- 6. Malo a Mizinda
- 7. Kinetics Dataset
- 8. CelebMask-HQ
- 9. Penn Treebank
- 10. VoxCeleb
- 11. SIXray
- 12. Ngozi zaku US
- 13. Kuzindikira Matenda a Ocular
- 14. Matenda a Mtima
- 15. CLEVR
- 16. Zodalira Padziko Lonse
- 17. KITTI – 360
- 18. MOT(Multiple Object Tracking)
- 19. PASCAL 3D+
- 20. Zitsanzo Zopunduka Pamaso pa Zinyama
- 21. MPII Human Post Dataset
- 22. UCF101
- 23. Audioset
- 24. Stanford Natural Language Inference
- 25. Kuyankha Mafunso Owona
- Kutsiliza
Masiku ano, ambiri aife timayang'ana kwambiri pakupanga makina ophunzirira makina ndi mitundu ya AI ndikuthana ndi zovuta pogwiritsa ntchito ma data apano. Koma choyamba, tiyenera kufotokozera deta, kufunikira kwake, ndi udindo wake pakupanga mayankho amphamvu a AI ndi ML.
Masiku ano, tili ndi unyinji wa nkhokwe zopezeka poyera zomwe tingachite kafukufuku kapena kupanga mapulogalamu othana ndi zovuta zenizeni m'magawo osiyanasiyana.
Komabe, kusowa kwa ma dataset apamwamba kwambiri kumabweretsa nkhawa. Deta yakwera kwambiri ndipo ipitilira kukula mwachangu mtsogolomu.
Mu positi iyi, tifotokoza zomwe zikupezeka kwaulere zomwe mungagwiritse ntchito popanga pulojekiti yotsatira ya AI.
1. CelebFaces Attributes Dataset
CelebFaces Attributes Dataset (CelebA) ili ndi zithunzi zopitilira 200K zodziwika bwino komanso zofotokozera 40 pachithunzi chilichonse, zomwe zimapangitsa kuti chikhale poyambira bwino kwambiri pama projekiti monga. kuzindikira nkhope, kuzindikira nkhope, chizindikiro (kapena mawonekedwe a nkhope) kutanthauzira, ndikusintha nkhope & kaphatikizidwe. Kuphatikiza apo, zithunzi zomwe zili mgululi zili ndi masinthidwe osiyanasiyana komanso zosokoneza.
2. DOTA
DOTA (Dataset of Kuzindikira Kwambiri mu Aerial Photos) ndi gulu lalikulu la data lozindikira zinthu lomwe lili ndi magawo 15 omwe amapezeka (monga, sitima, ndege, galimoto, ndi zina zotero), zithunzi 1411 zophunzitsidwa, ndi zithunzi 458 kuti zitsimikizidwe.
3. Google Facial Expression yofananira dataset
Poyerekeza ndi mawonekedwe a nkhope a Google ali ndi zithunzi pafupifupi 500,000, kuphatikiza zithunzi za nkhope 156,000. Ndizofunikira kudziwa kuti katatu kalikonse mu dataset iyi idafotokozedwa ndi anthu osachepera asanu ndi limodzi.
Deta iyi ndi yothandiza pama projekiti okhudza mawonekedwe a nkhope, monga kutengera zithunzi potengera mawu, kugawa zomwe akukhudzidwa, kaphatikizidwe ka mawu, ndi zina zotero. Kuti mupeze mwayi wofikira ku dataset, fomu yachidule iyenera kulembedwa.
4. Genome Yowonekera
Mayankho a Mafunso Owoneka M'malo osankha zambiri amapezeka mu Visual Genome. Zimapangidwa ndi zithunzi za 101,174 MSCOCO zokhala ndi ma 1.7 miliyoni a QA awiriawiri, ndi avareji ya mafunso 17 pa chithunzi chilichonse.
Poyerekeza ndi Visual Question Answering dataset, Visual Genome dataset ili ndi kugawa koyenera pakati pa mafunso asanu ndi limodzi: Chiyani, Kuti, Liti, Ndani, Chifukwa, ndi Motani.
Kuphatikiza apo, Visual Genome dataset imaphatikizanso zithunzi za 108K zomwe zidayikidwa kwambiri ndi zinthu, katundu, ndi kulumikizana.
5. LibriSpeech
LibriSpeech corpus ndi gulu la ma audiobook pafupifupi 1,000 kuchokera ku projekiti ya LibriVox. Ambiri mwa mabuku omvera amachokera ku Project Gutenberg.
Deta yophunzitsira imagawidwa m'magawo atatu a 100hr, 360hr, ndi 500hr sets, pomwe dev ndi test data ndi pafupifupi 5hr kutalika kwa audio.
6. The Cityspaces
Chimodzi mwazinthu zodziwika bwino kwambiri zamakanema a stereo okhala ndi mawonedwe akutawuni amatchedwa The Cityscapes.
Ndi mawu olondola a pixel omwe ali ndi malo a GPS, kutentha kwakunja, deta yoyenda bwino, komanso kawonedwe koyenera ka stereo, imaphatikizapo zojambulira zochokera kumizinda 50 yaku Germany.
7. Kinetics Dataset
Chimodzi mwazinthu zodziwika bwino zamakanema zozindikiritsa zochita za anthu pamlingo waukulu komanso wabwino kwambiri ndi dataset ya Kinetics. Pali mavidiyo osachepera 600 pagulu lililonse la maphunziro 600 a anthu, opitilira 500,000 onse.
Mafilimu adachotsedwa ku YouTube; Iliyonse imakhala ndi masekondi 10 ndipo ili ndi kalasi imodzi yokha yantchito yomwe yatchulidwa.
8. CelebMask-HQ
CelebAMask-HQ ndi mndandanda wa zithunzi 30,000 zamaso zowoneka bwino zokhala ndi masks odziwika bwino ndi makalasi 19 omwe amaphatikiza zinthu za nkhope monga khungu, mphuno, maso, nsabwe, makutu, pakamwa, milomo, tsitsi, chipewa, galasi lamaso, ndolo, mkanda, khosi, zinthu.
Setiyi ikhoza kugwiritsidwa ntchito kuyesa ndi kuphunzitsa kuzindikira nkhope, kuyang'ana nkhope, ndi ma GAN popanga ma algorithms opangira nkhope ndikusintha.
9. Penn Treebank
Chimodzi mwazinthu zodziwika bwino komanso zomwe zimagwiritsidwa ntchito nthawi zambiri powunika zitsanzo zama tagi otsatizana ndi corpus ya English Penn Treebank (PTB), makamaka gawo la corpus lolingana ndi zolemba za Wall Street Journal.
Liwu lirilonse liyenera kukhala ndi gawo lake la mawu olembedwa ngati gawo la ntchitoyo. Makhalidwe ndi mulingo wa mawu chinenero chitsanzo nthawi zambiri amagwiritsa ntchito corpus.
10. VoxCeleb
VoxCeleb ndi gulu lalikulu lozindikiritsa mawu lomwe limapangidwa kuchokera open-source media. VoxCeleb ili ndi mawu opitilira miliyoni miliyoni kuchokera kwa olankhula oposa 6k.
Monga momwe deta ikuphatikiza zomvera, zitha kugwiritsidwa ntchito pazowonjezera zosiyanasiyana, kuphatikiza kaphatikizidwe ka mawu, kulekanitsa mawu, kusamutsa kumaso kupita ku liwu kapena mosemphanitsa, ndikuphunzitsa kuzindikira nkhope kuchokera pavidiyo kuti zithandizire kuzindikira nkhope yapano. ma datasets.
11. SIXray
Dongosolo la SIXray limaphatikizapo zithunzi za X-ray 1,059,231 zomwe zasonkhanitsidwa kuchokera kumasiteshoni zasitima yapansi panthaka ndikufotokozedwa ndi oyang'anira chitetezo chaanthu kuti azindikire mitundu isanu ndi umodzi ya zinthu zoletsedwa: mfuti, mipeni, ma wrenches, pliers, lumo, ndi nyundo. Kuphatikiza apo, mabokosi omangirira a chinthu chilichonse choletsedwa awonjezedwa pamanja pamaseti oyesera kuti awunikire momwe zinthu zimayendera.
12. Ngozi zaku US
Zomwe polojekitiyi yapanga zawululidwa kale ndi dzina la dataset, ngozi za US. Zolemba izi zokhudza ngozi zapamsewu zapadziko lonse lapansi zikuphatikiza zambiri kuyambira February 2016 mpaka Disembala 2021 ndipo zikuphatikiza zigawo 49 ku USA.
Pafupifupi zolemba zangozi 1.5 miliyoni zilipo m'gululi. Zinasonkhanitsidwa munthawi yeniyeni pogwiritsa ntchito ma API angapo apamsewu.
Ma APIwa amafalitsa zidziwitso zamagalimoto zomwe zasonkhanitsidwa kuchokera kumadera osiyanasiyana, kuphatikiza makamera apamsewu, mabungwe azamalamulo, ndi maofesi amayendedwe aku US ndi maboma.
13. Kuzindikira Matenda a Ocular
Malo osungira ophthalmic ophthalmic database Ocular Disease Intelligent Recognition (ODIR) ali ndi chidziwitso cha odwala 5,000, kuphatikiza zaka zawo, mtundu wa fundus m'maso awo akumanzere ndi kumanja, komanso mawu osakira azachipatala.
Deta iyi ndi mndandanda weniweni wa deta ya odwala kuchokera kuzipatala zosiyanasiyana ndi zipatala ku China zomwe Shanggong Medical Technology Co., Ltd. Ndi kasamalidwe kaubwino, ndemanga zinalembedwa ndi anthu aluso owerenga.
14. Matenda a Mtima
Deta iyi ya matenda a Mtima imathandiza kuzindikira kukhalapo kwa matenda a mtima mwa wodwala malinga ndi magawo a 76 monga zaka, jenda, mtundu wa ululu pachifuwa, kupuma kwa magazi, ndi zina zotero.
Ndi milandu 303, nkhokweyo ikufuna kungosiyanitsa kukhalapo kwa matenda (mtengo 1,2,3,4) ndi kusakhalapo kwake (mtengo 0).
15. Mtengo CLEVR
Dongosolo la data la CLEVR (Compositional Language ndi Elementary Visual Reasoning) limatsanzira Mayankho a Mafunso Owoneka. Muli ndi zithunzi za zinthu zopangidwa ndi 3D, chithunzi chilichonse chikutsatiridwa ndi mafunso angapo olembedwa bwino omwe amagawidwa m'magulu angapo.
Pazithunzi zonse zamasitima ndi zotsimikizira ndi mafunso, gululi lili ndi zithunzi 70,000 ndi mafunso 700,000 ophunzitsidwa, zithunzi 15,000 ndi mafunso 150,000 kuti atsimikizidwe, ndi zithunzi 15,000 ndi mafunso 150,000 oyesa zinthu, mayankho, ma grafu ogwira ntchito, ndi magalasi ogwira ntchito.
16. Universal Dependencies
Pulojekiti ya Universal Dependencies (UD) ikufuna kupanga morphology yamitundu yosiyanasiyana komanso mawu ofotokozera amtundu wamitengo m'zilankhulo zambiri. Mtundu wa 2.7, womwe unatulutsidwa mu 2020, uli ndi mabanki amitengo 183 m'zinenero 104.
Mawuwa amapangidwa ndi ma tag a POW onse, mitu yodalira, ndi zilembo zapadziko lonse lapansi zodalira.
17. KITTI - 360
Imodzi mwama dataset omwe amagwiritsidwa ntchito kwambiri pama robot am'manja ndi kuyendetsa pawokha ndi KITTI (Karlsruhe Institute of Technology ndi Toyota Technological Institute).
Zimapangidwa ndi kuchuluka kwa magalimoto omwe adagwidwa pogwiritsa ntchito njira zingapo zamasensa, monga ma RGB apamwamba kwambiri, makamera a grayscale stereo, ndi makamera a 3D laser scanner. Zosungirako zasinthidwa pakapita nthawi ndi ofufuza angapo omwe adalemba pamanja magawo osiyanasiyana kuti agwirizane ndi zosowa zawo.
18. MOT(Multiple Object Tracking)
MOT (Multiple Object Tracking) ndi dataset yotsata zinthu zingapo zomwe zimaphatikizapo zochitika zamkati ndi zakunja za malo omwe pali anthu ambiri omwe amaphatikiza oyenda pansi ngati zinthu zomwe zimakonda. Kanema wa chochitika chilichonse agawidwa m'zidutswa ziwiri, imodzi yophunzitsira ndi ina yoyesa.
Dataset ikuphatikizapo kuzindikira zinthu m'mafelemu amakanema pogwiritsa ntchito zowunikira zitatu: SDP, Faster-RCNN, ndi DPM.
19. PASCAL 3D+
Pascal3D + multiview dataset imapangidwa ndi zithunzi zomwe zimasonkhanitsidwa kuthengo, mwachitsanzo, zithunzi zamagulu azinthu zomwe zimakhala ndi kusiyana kwakukulu, zojambulidwa muzochitika zosalamulirika, m'malo odzaza anthu, komanso m'malo osiyanasiyana. Pascal3D+ imaphatikizapo magulu 12 azinthu zolimba zotengedwa kuchokera ku dataset ya PASCAL VOC 2012.
Zinthuzi zili ndi chidziwitso cha mawonekedwe (azimuth, kukwera, ndi mtunda wa kamera). Pascal3D+ imaphatikizansopo zithunzi zojambulidwa kuchokera pagulu la ImageNet m'magulu 12 awa.
20. Zitsanzo Zopunduka Pankhope za Zinyama
Cholinga cha pulojekiti ya Facial Deformable Models of Animals (FDMA) ndi kutsutsa njira zomwe zilipo pakali pano pozindikiritsa ndi kutsata zizindikiro za nkhope ya munthu ndikupanga njira zatsopano zomwe zingathe kuthana ndi kusiyana kwakukulu komwe kuli ndi mawonekedwe a nkhope ya nyama.
Ma algorithms a pulojekitiyi adawonetsa kuthekera kozindikira ndikutsata malo omwe ali pankhope za anthu pomwe akulimbana ndi kusiyanasiyana komwe kumabwera chifukwa cha kusintha kwa nkhope kapena malo, kutsekeka pang'ono, ndi kuyatsa.
21. MPII Human Post Dataset
MPII Human Pose Dataset ili ndi zithunzi pafupifupi 25K, 15K zomwe ndi zitsanzo zophunzitsira, 3K zomwe ndi zitsanzo zovomerezeka, ndi 7K zomwe ndi zitsanzo zoyesa.
Maudindowa amalembedwa pamanja zolumikizana mpaka 16, ndipo zithunzizo zimatengedwa m'mafilimu a YouTube okhudza zochitika 410 zosiyanasiyana za anthu.
22. UCF101
Dongosolo la UCF101 lili ndi makanema 13,320 opangidwa m'magulu 101. Magulu 101 amenewa agawidwa m’magulu asanu: mayendedwe a thupi, kugwirizana kwa munthu ndi munthu, kuchita zinthu ndi anthu, kuimba zida zoimbira, ndi masewera.
Makanemawa akuchokera ku YouTube ndipo amakhala ndi maola 27.
23. Audioset
Audioset ndi gulu lazomvera lomwe limapangidwa ndi magawo 2 miliyoni ofotokozera anthu pamasekondi 10. Kufotokozera zambiri izi, akatswiri a ontology omwe ali ndi mitundu 632 ya zochitika amagwiritsidwa ntchito, zomwe zikutanthauza kuti mawu omwewo amatha kulembedwa mosiyana.
24. Stanford Natural Language Inference
Dongosolo la data la SNLI (Stanford Natural Language Inference) lili ndi ziganizo zokwana 570k zomwe zasanjidwa pamanja monga zotsutsana, zotsutsana, kapena zandale.
Malo ndi mafotokozedwe a zithunzi za Flickr30k, pomwe zongopeka zidapangidwa ndi ofotokozera omwe ali ndi unyinji omwe adapatsidwa maziko ndikulangizidwa kuti apereke ziganizo zomveka, zotsutsana, komanso zandale.
25. Kuyankha Mafunso Owoneka
Visual Question Answering (VQA) ndi gulu lomwe lili ndi mafunso opanda mayankho okhudza zithunzi. Kuti muyankhe mafunsowa, muyenera kumvetsetsa masomphenya, chinenero, ndi luntha.
Kutsiliza
Pamene kuphunzira pamakina ndi luntha lochita kupanga (AI) kukuchulukirachulukira mu bizinesi iliyonse komanso m'moyo wathu watsiku ndi tsiku, momwemonso kuchuluka kwazinthu ndi chidziwitso chomwe chilipo pankhaniyi.
Zosungidwa zapagulu zomwe zakonzedwa zimapereka poyambira zabwino kwambiri zopangira mitundu ya AI komanso kulola opanga mapulogalamu a ML kuti asunge nthawi ndikuyang'ana pazinthu zina zamapulojekiti awo.
Siyani Mumakonda