Table of Contents[Hūnā][Hōʻike]
- 1. ʻIkepili ʻikepili ʻano ʻano CelebFaces
- 2. DOTA
- 3. Google Facial Expression hoʻohālikelike ʻikepili
- 4. ʻIkepili Genome
- 5. Haʻiʻōlelo Libri
- 6. Nā Kūlanakauhale
- 7. Kinetics Dataset
- 8. CelebAMask-HQ
- 9. Penn Treebank
- 10. VoxCeleb
- 11. SIXray
- 12. ʻAmelika Hui Pū ʻIa
- 13. Ka ʻike ʻana i ka maʻi maka
- 14. maʻi maʻi puʻuwai
- 15. CLEVR
- 16. Nā Kūlana Nui
- 17. KITI – 360
- 18. MOT
- 19. PASCAL 3D+
- 20. Nā Kiʻi Hoʻohālikelike Maka o nā Holoholona
- 21. MPII Human Post Dataset
- 22. UCF101
- 23. Audioset
- 24. Ka Manaʻo ʻŌlelo Kūlohelohe ʻo Stanford
- 25. Pane Ninau Makani
- Panina
I kēia mau lā, ʻo ka hapa nui o mākou e kālele ana i ka hoʻomohala ʻana i ke aʻo ʻana i ka mīkini a me nā hiʻohiʻona AI a me ka hoʻoponopono ʻana i nā pilikia me ka hoʻohana ʻana i nā ʻikepili o kēia manawa. Akā ʻo ka mea mua, pono mākou e wehewehe i kahi dataset, kona koʻikoʻi, a me kāna kuleana i ka hoʻomohala ʻana i nā hoʻonā AI a me ML ikaika.
I kēia lā, loaʻa iā mākou he plethora o ka wehe ʻana i ka ʻikepili kumu e hana ai i ka noiʻi a i ʻole ka hoʻomohala ʻana i nā noi e hoʻoponopono i nā pilikia o ka honua maoli i nā ʻano ʻāpana like ʻole.
Eia naʻe, ʻo ka liʻiliʻi o nā ʻikepili helu kiʻekiʻe ke kumu o ka hopohopo. Ua piʻi nui ka ʻikepili a e hoʻomau i ka hoʻonui ʻana i ka wikiwiki i ka wā e hiki mai ana.
Ma kēia pou, e uhi mākou i nā ʻikepili i loaʻa i hiki iā ʻoe ke hoʻohana e hoʻomohala i kāu papahana AI aʻe.
1. ʻIkepili ʻikepili ʻano ʻano CelebFaces
Loaʻa iā CelebFaces Attributes Dataset (CelebA) ma luna o 200K mau kiʻi kaulana a me 40 mau hōʻike hōʻike no kēlā me kēia kiʻi, e lilo ia i wahi hoʻomaka maikaʻi loa no nā papahana e like me ʻike maka, ka ʻike maka, ka ʻāina ʻāina (a i ʻole ka ʻaoʻao o ka maka) ka hoʻonohonoho ʻana, a me ka hoʻoponopono ʻana a me ka synthesis. Eia kekahi, ʻo nā kiʻi i loko o kēia hōʻiliʻili i loaʻa i kahi ākea o nā ʻano kūlana a me nā clutter backdrop.
2. DOTA
DOTA (Ka ʻikepili o ʻIke Mea ma Aerial Photos) he waihona helu nui no ka ʻike ʻana i nā mea i loaʻa he 15 mau ʻano maʻamau (e laʻa, moku, mokulele, kaʻa, etc.), 1411 kiʻi no ke aʻo ʻana, a me 458 kiʻi no ka hōʻoia.
3. ʻO ka ʻikepili hoʻohālikelike Google Facial Expression
Aia ma kahi o 500,000 mau kiʻi triplets ka Google hoʻohālikelike hoʻohālikelike, me 156,000 mau kiʻi maka. Pono e hoʻomaopopo ʻia ua hōʻike ʻia kēlā me kēia triplet i kēia ʻikepili e nā mea helu kanaka ma lalo o ʻeono.
He mea pono kēia waihona no nā papahana e pili ana i ka nānā ʻana i ka ʻike maka, e like me ke kiʻi kiʻi ma muli o ka hōʻike, ka hoʻokaʻawale ʻana i nā manaʻo, ka hoʻohui ʻana i ka ʻōlelo, a pēlā aku. No ka loaʻa ʻana o ka ʻikepili, pono e hoʻopiha i kahi palapala pōkole.
4. ʻIkepili Genome
Loaʻa ka ʻikepili pane pane ʻike ma kahi kaiapuni koho nui ma Visual Genome. He 101,174 mau kiʻi MSCOCO me 1.7 miliona mau hoa QA, me ka awelika o 17 mau nīnau no ke kiʻi.
I ka hoʻohālikelike ʻana i ka ʻikepili pane pane ʻana i ka nīnau nīnau, ʻoi aku ka maikaʻi o ka hoʻolaha ʻana o ka dataset Visual Genome ma nā ʻano nīnau ʻeono: He aha, ma hea, i ka manawa, ʻo wai, no ke aha, a pehea.
Eia kekahi, ʻo ka Visual Genome dataset nā kiʻi 108K i hoʻopaʻa nui ʻia me nā mea, nā waiwai, a me nā pilina.
5. LibriSpeech
ʻO ka LibriSpeech corpus kahi hōʻiliʻili o nā hola 1,000 o nā puke leo mai ka papahana LibriVox. ʻO ka hapa nui o nā puke leo mai Project Gutenberg.
Hoʻokaʻawale ʻia ka ʻikepili aʻo i ʻekolu mau ʻāpana o 100hr, 360hr, a me 500hr sets, aʻo ka dev a me ka ʻikepili hōʻike ma kahi o 5hr i ka lōʻihi o ka leo.
6. Nā Kūlanakauhale
ʻO kekahi o nā ʻikepili nui kaulana loa o nā wikiō stereo me nā hiʻohiʻona kūlanakauhale i kapa ʻia ʻo The Cityscapes.
Me nā hōʻike kikoʻī pololei e pili ana i nā wahi GPS, ka mahana o waho, ka ʻikepili ego-motion, a me nā hiʻohiʻona stereo kūpono, e komo pū ana nā hoʻopaʻa leo mai 50 mau kūlanakauhale Kelemania ʻokoʻa.
7. ʻIkepili Kinetics
ʻO kekahi o nā waihona wikiō kaulana loa no ka ʻike ʻana i ka hana a ke kanaka ma ke ʻano nui a me ka maikaʻi maikaʻi ʻo ka Kinetics dataset. Aia ma kahi o 600 mau wikiō wikiō no kēlā me kēia o nā papa hana kanaka 600, ʻoi aku ka nui o 500,000.
Ua huki ʻia nā kiʻiʻoniʻoni mai YouTube; ʻO kēlā me kēia ma kahi o 10 kekona ka lōʻihi a hoʻokahi wale nō papa hana i helu ʻia.
8. CelebAMask-HQ
ʻO CelebAMask-HQ kahi hōʻiliʻili o 30,000 mau kiʻi maka kiʻekiʻe me nā masks annotated a me 19 mau papa e loaʻa ai nā ʻāpana helehelena e like me ka ʻili, ka ihu, nā maka, nā lae, nā pepeiao, ka waha, ka lehelehe, ka lauoho, ka pāpale, ka aniani, ka apo pepeiao, ka lei, ʻāʻī, mea.
Hiki ke hoʻohana ʻia ka ʻikepili no ka hoʻāʻo ʻana a hoʻomaʻamaʻa i ka ʻike maka, ka hoʻopili ʻana i ke alo, a me nā GANs no ka hoʻokumu ʻana i ka maka a me ka hoʻoponopono algorithms.
9. Penn Treebank
ʻO kekahi o nā hui kaulana a hoʻohana pinepine ʻia no ka loiloi ʻana i nā hiʻohiʻona no ka hoʻopili ʻana i ke kaʻina o ka English Penn Treebank (PTB) corpus, ʻo ia hoʻi ka ʻāpana o ke kino e pili ana i nā ʻatikala Wall Street Journal.
Pono kēlā me kēia huaʻōlelo i kā lākou ʻāpana o ka haʻiʻōlelo i kāʻei ʻia ma ke ʻano he ʻāpana o ka hana. Kaulana ʻano a me ka pae ʻōlelo hoʻohālike ʻōlelo hoʻohana pinepine i ke kino.
10. VoxCeleb
ʻO VoxCeleb kahi waihona ʻike ʻōlelo nui i hana ʻia mai pāʻoihana wehe-puna. Loaʻa iā VoxCeleb ma luna o hoʻokahi miliona mau ʻōlelo mai ka 6k mau mea ʻōlelo.
No ka loaʻa ʻana o ka ʻikepili i ka leo-kiʻi, hiki ke hoʻohana ʻia no nā ʻano noiʻi ʻē aʻe, e like me ka synthesis haʻiʻōlelo ʻike, ka hoʻokaʻawale ʻana i ka haʻiʻōlelo, ka hoʻololi ʻana i ke ʻano o ke ʻano mai ke alo a i ka leo a i ʻole, a me ke aʻo ʻana i ka ʻike maka mai ka wikiō e hoʻohui i ka ʻike maka o kēia manawa. nā papa helu.
11. SIXray
Aia ka SIXray dataset i nā kiʻi X-ray 1,059,231 i hōʻiliʻili ʻia mai nā kahua kaʻaahi a i hōʻike ʻia e nā mea nānā palekana kanaka e ʻike i ʻeono ʻano nui o nā mea i pāpā ʻia: nā pū, nā pahi, nā wrenches, pliers, scissors, a me nā hāmare. Eia kekahi, ua hoʻohui lima ʻia nā pahu hoʻopaʻa no kēlā me kēia mea i ʻae ʻole ʻia i nā hoʻonohonoho hoʻāʻo i mea e loiloi ai i ka hana o ka localization mea.
12. ʻAmelika Hui Pū ʻIa
Ua hōʻike mua ʻia ka waiwai o ka papahana e ka inoa o ka dataset, US Accidents. Aia kēia ʻikepili i nā ulia kaʻa holoʻokoʻa i ka ʻike mai Pepeluali 2016 a hiki i Dekemaba 2021 a uhi ʻia nā mokuʻāina he 49 ma USA.
Aia ma kahi o 1.5 miliona mau moʻolelo ulia pōpilikia i kēia manawa i kēia hōʻiliʻili. Ua hōʻiliʻili ʻia i ka manawa maoli me ka hoʻohana ʻana i kekahi mau API traffic.
Hoʻouna kēia mau API i ka ʻike kaʻa kaʻa i hōʻiliʻili ʻia mai nā kumu like ʻole, e komo pū ana me nā kāmela kaʻa, nā hui hoʻokō kānāwai, a me ka US a me nā keʻena kaʻa mokuʻāina.
13. ʻIke ʻana i ka maʻi ʻocular
Aia ka ʻikepili ophthalmic i hoʻonohonoho ʻia ʻo Ocular Disease Intelligent Recognition (ODIR) i ka ʻike e pili ana i 5,000 mau maʻi, me ko lākou mau makahiki, ke kala o ka waihona kālā ma ko lākou mau maka hema a ʻākau, a me nā huaʻōlelo diagnostic a ka poʻe lapaʻau.
He hōʻiliʻili maoli kēia o nā ʻikepili maʻi mai nā halemai like ʻole a me nā keʻena lapaʻau ma Kina i loaʻa iā Shanggong Medical Technology Co., Ltd. Me hooponopono hooponopono maikai, hōʻailona ʻia e nā kānaka heluhelu akamai.
14. ʻO ka maʻi maʻi
Kōkua kēia ʻikepili maʻi puʻuwai i ka ʻike ʻana i ke ola ʻana o ka maʻi puʻuwai i ka mea maʻi ma muli o 76 mau palena e like me ka makahiki, ke kāne, ke ʻano ʻeha o ka umauma, hoʻomaha ke koko, a pēlā aku.
Me nā hihia 303, ʻimi ka ʻikepili e hoʻokaʻawale i ke ʻano o kahi maʻi (waiwai 1,2,3,4) mai kona nele (waiwai 0).
15. CLEVR
Hoʻohālike ka ʻikepili CLEVR (Compositional Language and Elementary Visual Reasoning) i ka pane ʻana i nā nīnau. Loaʻa iā ia nā kiʻi o nā mea 3D-rendered, me kēlā me kēia kiʻi i hui pū ʻia me kahi moʻo o nā nīnau haku mele i hoʻokaʻawale ʻia i kekahi mau ʻano.
No nā kiʻi kaʻaahi a me ka hōʻoia ʻana a me nā nīnau, loaʻa nā kiʻi 70,000 a me 700,000 mau nīnau no ka hoʻomaʻamaʻa ʻana, 15,000 kiʻi a me 150,000 mau nīnau no ka hōʻoia ʻana, a me 15,000 mau kiʻi a me 150,000 nīnau no ka hoʻāʻo ʻana e pili ana i nā mea, nā pane, nā kiʻi hiʻohiʻona hana, a me nā kiʻi hana.
16. Nā Kaulike Universal
Ke manaʻo nei ka papahana Universal Dependencies (UD) e hana i ka morphology cross-linguistically uniform a me syntax treebank annotation no nā ʻōlelo he nui. ʻO ka version 2.7, i hoʻokuʻu ʻia i ka makahiki 2020, he 183 mau waihona lāʻau ma 104 mau ʻōlelo.
Hoʻokumu ʻia ka hōʻailona me nā inoa POW āpau, nā poʻo hilinaʻi, a me nā lepili hilinaʻi honua.
17. KITI – 360
ʻO kekahi o nā ʻikepili i hoʻohana pinepine ʻia no nā robots mobile a kalaiwa kū kaʻawale ʻO KITTI (Karlsruhe Institute of Technology a me Toyota Technological Institute).
Hoʻokumu ʻia i nā hiʻohiʻona kaʻa o nā hola i hopu ʻia me ka hoʻohana ʻana i nā ʻano o nā ʻano sensor, e like me ka RGB kiʻekiʻe, ka stereo grayscale, a me nā kāmeʻa scanner laser 3D. Ua hoʻomaikaʻi ʻia ka ʻikepili i ka wā lōʻihi e kekahi mau mea noiʻi i hōʻike lima lima i nā ʻāpana like ʻole o ia mea e kūpono i ko lākou pono.
18. MOT
ʻO MOT (Multiple Object Tracking) he papa helu no ka huli ʻana i nā mea he nui e komo pū ana me nā hiʻohiʻona o loko a me waho o nā wahi lehulehu e komo pū ana ka poʻe hele wāwae i nā mea hoihoi. Wehe ʻia ke wikiō o kēlā me kēia hiʻohiʻona i ʻelua ʻāpana, hoʻokahi no ka hoʻomaʻamaʻa ʻana a ʻo kekahi no ka hoʻāʻo ʻana.
Aia i loko o ka waihona mea ike i nā kiʻi wikiō e hoʻohana ana i ʻekolu mea ʻike: SDP, Faster-RCNN, a me DPM.
19. PASCAL 3D+
ʻO ka Pascal3D+ multi-view dataset ka mea i hana ʻia me nā kiʻi i hōʻiliʻili ʻia ma ka nahele, ʻo ia hoʻi, nā kiʻi o nā ʻano mea me ka nui o nā ʻano like ʻole, i hopu ʻia i nā kūlana i kāohi ʻole ʻia, ma nā wahi ākea, a ma nā kūlana like ʻole. Loaʻa iā Pascal3D+ he 12 waeʻano mea ʻoʻoleʻa i huki ʻia mai ka waihona helu PASCAL VOC 2012.
Loaʻa kēia mau mea i ka ʻike posture i kaha ʻia ma luna o lākou (azimuth, kiʻekiʻe, a me ka mamao i ke kamera). Hoʻokomo pū ʻia ʻo Pascal3D+ i nā kiʻi pose-annotated mai ka hōʻiliʻili ImageNet i kēia mau ʻāpana 12.
20. Nā ʻano hoʻohālike o ka helehelena o nā holoholona
ʻO ka pahuhopu o ka papahana Facial Deformable Models of Animals (FDMA) ʻo ia ka hoʻohālikelike ʻana i nā ʻano hana i kēia manawa i ka ʻike ʻana a me ka nānā ʻana i ka ʻāina kanaka a me ka hoʻomohala ʻana i nā algorithms hou e hiki ai ke hoʻoponopono i ka ʻano like ʻole o ka helehelena holoholona.
Ua hōʻike nā algorithms o ka papahana i ka hiki ke ʻike a nānā i nā hiʻohiʻona ma nā maka o ke kanaka i ka wā e pili ana i nā ʻano like ʻole i hoʻohuli ʻia e nā loli i nā manaʻo a i ʻole nā kūlana o ka helehelena, nā ʻāpana ʻāpana, a me nā kukui.
21. MPII Human Post Dataset
Aia i loko o ka MPII Human Pose Dataset ma kahi o 25K kiʻi, 15K o ia mau laʻana hoʻomaʻamaʻa, 3K o ia mau hōʻailona hōʻoia, a he 7K o ia mau hōʻike hoʻāʻo.
Hoʻopaʻa lima ʻia nā kūlana me 16 mau hui kino, a lawe ʻia nā kiʻi mai nā kiʻiʻoniʻoni YouTube e uhi ana i 410 mau hana kanaka.
22. UCF101
Aia i loko o ka waihona UCF101 he 13,320 wikiō i hoʻonohonoho ʻia i 101 mau ʻāpana. Hoʻokaʻawale ʻia kēia mau ʻāpana he 101 i ʻelima mau ʻāpana: ka neʻe ʻana o ke kino, ka pilina kanaka-kanaka, ka pilina kanaka-mea, ka hoʻokani pila, a me nā haʻuki.
No YouTube nā wikiō a he 27 mau hola ka lōʻihi.
23. Audioset
ʻO Audioset kahi hōʻike leo hanana i hana ʻia ma luna o 2 miliona mau ʻāpana wikiō 10-kekona i hōʻike ʻia e ke kanaka. No ka hōʻike ʻana i kēia ʻikepili, hoʻohana ʻia kahi hierarchical ontology me 632 mau ʻano hanana, ʻo ia ka manaʻo e hōʻailona ʻokoʻa ke kani like.
24. Manaʻo ʻŌlelo Kūlohelohe ʻo Stanford
Aia i loko o ka waihona SNLI (Stanford Natural Language Inference) he 570k mau huaʻōlelo i hoʻokaʻawale lima ʻia ma ke ʻano he entailment, condiction, a i ʻole neutral.
ʻO nā wahi nā wehewehe kiʻi Flickr30k, ʻoiai ua hoʻomohala ʻia nā kuhiakau e nā annotators i hoʻokumu ʻia e ka lehulehu i hāʻawi ʻia i kahi kumu a kauoha ʻia e hana i nā ʻōlelo entailing, kū'ē, a kū ʻole.
25. Pane Ninau Makani
ʻO ka Pane Nīnau Kiʻi (VQA) kahi waihona i loaʻa nā nīnau wehe e pili ana i nā kiʻi. No ka pane ʻana i kēia mau nīnau, pono ʻoe e ʻike i ka ʻike, ka ʻōlelo, a me ka noʻonoʻo.
Panina
I ka lilo ʻana o ke aʻo ʻana i nā mīkini a me ka naʻauao artificial (AI) ma nā ʻoihana āpau a i ko mākou ola i kēlā me kēia lā, pēlā nō ka nui o nā kumuwaiwai a me nā ʻike i loaʻa ma ke kumuhana.
Hāʻawi nā ʻikepili lehulehu i mākaukau i kahi hoʻomaka maikaʻi no ka hoʻomohala ʻana i nā hiʻohiʻona AI me ka ʻae pū ʻana i nā polokalamu polokalamu ML manawa e mālama i ka manawa a nānā i nā mea ʻē aʻe o kā lākou papahana.
Waiho i ka Reply