So I wanttotalk a littlebitaboutscheduleandstuffandwhat's happeningonwhatyoucanexpect.
Butbefore I doanyofthatinorderforyoutofallalongwiththemaincontent, thatisthecontentoftoday's lifestream, whichis a veryspecialexcitedcontent.
Veryexcitedaboutyoudoneedtodownloadsomesoftware.
So I'm justgonnatellyouwhattodownloadsto.
Megansaiditgoinginthebackgroundwhile I babble, babbleonandonaboutotherstuff.
So I'm gonnaopenup a Webbrowser, somethinglikethis.
Go, go, go, goUgly.
I don't knowhowtopronouncethatchrome a comb.
A.
I thinkit's pronouncedWebbrowser, and I'm goingtoaskyoutonavigateThioifyousochoose, umrunwaymldotcomSogotorunwaymldotcomVenueonewhoheadovertothislinkunderdownloadbeta.
Normally, I'm justinthisroomtalkingtomyself, but I amtalkingtoChrisValenzuela.
How'd I doonthepronunciation?
Prettygood.
OneofthefoundersandcreatorsofRunway, whoishereinthechatSayhitoChrisinthechattoansweryourquestionsandtocontrolmyquestionsas I fumbledmywaythroughtryingtolearnaboutrunway.
So I'm gonnacomebacktotalkaboutwhatrunwayisandwhat I'm plantodointoday's livestream.
But I justwantedtoatleastsetyouonThio, downloadingthisifyouwantto, and I'm goingtogiveyou a couponcodein a littlebittogetsomeextracreditsaswell.
Sobutbefore I dothat, Sowhyyou'redownloadingthat?
What's happeninghere?
Whereareyou?
Wheream I?
Wherearewewithwhatisthat?
What's whatisthemeaningoftheuniverse?
Thatisnot a question I willanswer.
Sothisis a little, uh, neighborhoodYouTubechannelcalledtheCodingTrain.
Andeverytime I saycodingtrain, I havetoblow a trainwhistle.
It's inthiscontractorsigned a whileagoand, um, where I makecomputerprogrammingtutorialsandotherwisewastetimereadingthingslikerandomnumbers.
Andmostofmyactivityhappensduringtheschoolyearbecause I alsoworkat a schoolNewYorkUniversityteaching a programcalled T P.
Andalsothat's a graduateprogramandundergraduateprogramcalled I Am A.
They'rebothpartofSchooloftheArts, sosummeris a littlebitlighterformeintermsofcontentanyway.
Butthissummer's airparticularlyuniquesummerinthatthisroomthat I'm standinginrightnowisbeingcloseddown, and I'm movingthecamerasandequipmentto a newroomonourintheBrooklyncampussometimethismonthornextmonth.
So I'm gonnalosesometimewhere I don't have a recordingstudioandwe'llbesettingup a newrecordingstudio, whichhopefullysomebetterstuffandbackfullforceinSeptember.
However, therearemanyvideosstilloutstandingthat I amworkingoneditingtogetherandreleasingnumberoneisandthisisveryexciting a fulldocumentationoffvideoaboutalloftheequipmentandsoftwareinthisroomthat I'm using.
Sothis, uh, codingtrainstudiowhichhasbeenheresince I don't knowwhen I firststarted, I gottalookbackandfigureoutwhatwasthefirstlivestream I didfromthisroom.
Thefirstlivestreamonthecoatingtrain, I think, was 2015 orperhapsearlierthanthat.
But I was I wasdoingeverythingfrommyofficedownstairsfor a whileanyway.
That's notsuperimportant.
Um, sothat's comingout.
I have a twopartseriesonhowtoprogramyourownprocessingJavalibrary, which I knoweverybodyontheInternetisjustwaiting.
That's gonnablowupYouTube.
That's gonnabe, likehighestviewedvideoever, where I showyouhowtobuild a job, a libraryineclipsebecausethat's likealltherage.
Nowthat's whateverisdoing, right?
Right?
Okay, take a deepbreathherefor a second.
That's comingoutofthat.
Theeditedversionoftheticktacktoecodingchallengethat I didon a previouslivestreamiscoming, andalsoif I couldgetmyacttogetherbecause I forgottorecordsomestuffforthattoday, a editedversionoftheEuroevolutionsteering, carracingcodingchallengethings.
Sothat's alsocoming.
Sothere's a bunchofcontentthatwillbecomingoutinthechannel.
ButthismightbethelastlivestreamuntilSeptember, or I mightpopbackfor a surpriselivestreamlikethisoneonsometopic, likerunway.
Okay, now I'm lookingatthecheck.
Nowlet's sayyouhappentobesupportingthecodingtrainas a memberonYouTubeor a patrononpatron, and I think I amgoingtoputtingthisoutthere.
Theuniversity.
I'm notsureaboutit, but I mightshutdownthePatrionandmoveoverentirelytoYouTubememberships.
I thinkif I goheretoYouTubeslashthecodingtrainslashJoin?
Um, it'lltellme.
Doesn't tellmeanything.
No, mebecause I'm already a memberbecause I'm notloggedin.
Whoknows?
But, look, I'm livecommawaving.
CanthoseMe?
Allright.
Um, soisokay.
Cookcookiecrumbisasking.
Isrunwayokayforbeginners?
Because I'm a beginner.
Yes.
I saidthatwaytooloudlywhenwesaidwe'reConley.
Yes.
So I thinkthereareontherunwayis a reallyinterestingthing, and I thinkthatitwouldbebestformetostartusingittoexplainit, butbecause I thinkthereareaspectsofitthatthewholesystem, byitsnatureofthewaythatit's builtas a pieceofsoftwareismeantforbeginners, butitis a toolthatallowsyoutoexploreprettydeeplytheworldofmachinelearningmodels, someofwhicharequitesophisticated.
Andyoucanwanderdown a lotofadvancedittrickyandcomplexrabbitholes.
So I thinkthere's a widerangehereofkindofunderstandingandrequirednotrequired, butthatthatispartofusingrunway.
Butwhat I liketodointhecodingtrainishaveeverything.
Butjustincaseitdoeshappentomakesensetodothatbecausethere's a lotofmebabbling, and I'm sure I'llwastesometimelaterchasingdownsomeweirdbuggerspieceofdocumentationorreadingrandomnumbers.
Butonewilljustbeanintroductiontorunwayas a platformandsigningupfor a runwayaccount, andthentheotherwillbemaybedoing a codingchallengewhere I trytogettheresults.
Theoutputofthemachinelearningmodelinto a P fivejazzschedule.
I'llgettothatlater.
Andreally, letmejustmentionifyouhavequestionsaboutRunwayaskedhiminthechatbecauseyoucouldhave a wholecompletelyseparateconversationgoingonwithoutme.
Um, allright, letmejustcoveroneotherthing, becausehowaroundisaskingDoyouhavetopaytokeepusingitSo I willgetintothisAs I lookatRunwayRunwayissomethingthatyoucanuse.
Youdon't needtoenter a creditcard.
Youcanuseitwithoutpayinganymoney.
However, thereissomecaveatstothis.
Theoneofthethingsthat I'llbedoingisrunningthemachinelearningmodelswithrunwayinthecloud, andthatrequirespaying a feebytheminuteorbythehourbytheminute.
ForcloudGPUusageeveryrunwayuserwhenyousignupyouautomaticallyget $10 offreecredits, andas I gothroughthesignupprocess, I'm gonnagiveyou a couponcode.
Youactuallydoquite a bitwith $20 credittobesurprised.
Andthentowardstheendofthelinestreet, maybewe'lltalkaboutwaysthatyoucanactuallyrunthemachinelearningmodelslocallyonyourlaptopifyou're a laptop, has a GPU.
Obviously, there's something.
There's quite a bitoflimitationtothat, dependingonwhatcomputeryou'reusing.
Butpress a buttoncalledInstall, anditgivesyouaninterfacetoplaywiththosemodels, experimentwiththosemodelsandthenbroadcasttheresultsofthosemodelstosomeotherpieceofsoftware.
I amgoingtopokethroughthemandshowyouhowtheyworkwithaneyetowardsatleastshowingyouhowtopairrunwaywithprocessingandhowtopoweronewaywith P fiveJsandalsoshowyouwherethere's lotsofotherexamplesandthingsyoucandowiththeotherplatformsandstufflikethat.
Youcouldcomehereandplayaroundwithit, butthere's there's a lotofyoucangoverydeeptoo, ifyouwant, findthepaper, readthroughthenotesandunderstandmoreaboutwhatthismodel, howitwasbuilt, whatdateitwastrainedon, whichisalways a veryimportantquestiontoaskwheneveryou'reusing a machinelearningmodel, so I canalsowecouldseetheirattributionishere.
Yourimagesegmentationis a commonenoughfeatureofmachinelearningmodelsthatRunwayhasbuiltintoitanentiredrawingenginesothatyoucanplayaroundwithimagesegmentationsoyoucouldseethesearethecolorsfordifferentlabels.
Somaybewhat I wantitlookslikeit's a lotoftransportationstuff.
Somaybewhat I wantislet's try.
Let's trydrawingsomepeople.
I'm gonnadothat.
Ifwedoitat a diversion, thiswillgetspedup.
Howam I doing?
It's drawingtopeople.
I don't thinkthisisactuallyhowpeoplelookin a reallyimage.
I wonderif I don't needtoDo I needtobeinterestingthingaboutjoining?
Doesitmatterif I doitinthisshapeoftheairplane, orisitjustgonnafigureitoutbecauseitknowswhattheshapeofanairplaneis?
That's actually a muchbetterairplanethan I imaginedmyselfdrawingwhatelse?
Umelephant.
Who?
Oh, boy.
There's a lotofotherstuff.
It's nottransportation, wine, glass.
It's gonnabe a verylargewineglasswithtwopeoplewith a airplaneand a wineglassflyingoverhead.
Okay, arewedoingnow?
I'm gonnachooseanoutput, and I justwanttodopreview, right, becausepreviewrightnow, it's like I'm notactually I don't needtoexportthis, andyouneedtouseitsomewhereelse.
I justwanttoplayaroundwithitinrunwayitself.
So I'm gonnapreviewand, uhandherecomesthemostNot.
Andnowhere's thething.
Okay.
Now I haveselectedmyinput, whichisjustthesegmentationinterfaceofrunwayitself.
I haveselectedmyoutput, whichisjust a preview.
Nowit's timeformetorunthemodel.
Andherewego.
Runremotelysoremote.
GPUenabled.
Andyoucouldseejustbysigningupforrunway, I have $10 inremoteGPUcredits.
Nowlet's putlikeanorangerightnexttoitforthecoupleorangestomake a littlebowloffruit.
Wow.
Thisiscrazy.
Wow.
Uh, where's mysearch?
Um, nowhearthis.
Let's seewhatelse?
Oneoftheirfruits.
Whatisthere?
A bananaisprobablybanana.
Okay, solet's put a banana.
Okay, this I gottastop.
That's prettyamazing.
Soagain, herewasjust a littlemomentlaterofbeingbeing a littlemorethoughtfultothinkabouthowthismodelactuallyworks.
Andwhatwhatmightknowif I knewIf I lookedatthedataset, whichisthefairlywellknown I imagineCocoImagedatasetThat's probablygoingthio.
GivemeevenMaurinformationtothinkaboutwhatit's goingtodowell, butyoucanseehowit's abletosortofthinkabout a sortoflittleseat a littlelittlepileoffruithereon a woodbackgroundonthis.
Anyquestionsthatpeoplewanttoaskinthechatthat I caneitheranswermyselforaskChrisabout?
Yes.
Andso, um, I noticedthatChrisjustwroteinthechat.
SomemodelsonlyworkonremoteGPU.
Soifyouwant I don't knowifthis, youknow, ifoffhand, ifthisisonethatwillworklocally?
No.
Sothisisonethatonlyworksonremote.
Cheap.
Youothermodelswereworkedlocally.
I think I'm gonnatrytorunposenet, whichcanrunlocally.
Andsoitwon't useanyGPUcreditsforthat.
Um, andthensomemodelswillworklocally, butrequireanotherinstallation, anotherdependency, whichissomethingcalledDocker, which I couldtalkaboutherinstallatsomepoint.
But I knowthatRunwayisworkingonmakingthatprocesseveneasiertobeabletorunstufflocally.
Um, okay.
Can I lowerresolutiontospeedup?
That's aninterestingquestion.
So I hope I canseeherethattheresolutionissetto 6 40 by 3 60 Isthatsomethingthatcanchangewiththismodel?
Good.
No, butthemodel I couldn't changetheinterface.
Yeah, so?
Sothisis a goodquestionbecauseyouknowthewaymostmachinelearningmodelsworkinparticular, ifthey'refixedinputsandfixedoutputs, notlike a kindofsequentialmodel.
Theoutputisfixed, theinputisfixed.
Requires a particularresolutionbecausetheneuralnetworkisexpecting a certainamountofnumberstocomein.
Soyoucandotosomeextentyoucoulddoresizingandresamplingif I gobacktomydiagramherelike I coulddosomeresizingandresamplinghere.
Or I coulddosomeresizingresamplingherebeforetopreparetheinputorafter I gettheoutput.
Butthemodelitself, theinputimagehastobe a particulardimension.
Andtheoutputisonlygonnacomeinthatdimension.
Somesomemodelsarereleased, I thinkwith, youknow, like a differentversion, doesn't itwithwhichmaybeoutputdifferentdimensions.
I think I'veseenthat.
Butinthisparticularone, it's fixeddimensions.
Okay.
Um, allright.
I couldn't, likeplaywiththisforever.
I couldplaywiththisforever, but I think I'vegot a knot.
Allright.
Umallright, sonowlet's letmethinkaboutthis.
Whatelsedo I wanttoshowherebefore I I I thinkwhat I'lldoistoseparate, like, sortofliketutorials.
But I don't thinkit's justinsortofthis, likegeneraltutorialabouttheplatformas a whole.
Whatelsedo I wanttosay?
Umsolet's lookatsoinputwise, I couldalso.
So I chosetodosegmentationhere, but I couldalsouse a file.
Soif I wantedtoopen a fileonthecomputer, I coulddoitthatwayandthenoutput.
If I changetoexport, I couldalsoactuallyexportthatthioofriot, differentformats.
Butofcourse I couldalsorightherejustunderpreview, I canclickthisdownloadsavebutton, andnow I amsavingforevermorethatthisparticularimageofthefilenowwhat's reallyimportanthere, actuallymoreimportanthereisundernetwork.
Soifwhat I wantedtodolet's clickoverhereunderNetwork. 00:52:46.390 -