?url_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Adc&rft.title=DiverseNet%3A+When+One+Right+Answer+is+not+Enough&rft.creator=Firman%2C+M&rft.creator=Campbell%2C+NDF&rft.creator=Agapito%2C+L&rft.creator=Brostow%2C+GJ&rft.description=Many+structured+prediction+tasks+in+machine+vision+have+a+collection+of+acceptable+answers%2C+instead+of+one+definitive+ground+truth+answer.+Segmentation+of+images%2C+for+example%2C+is+subject+to+human+labeling+bias.+Similarly%2C+there+are+multiple+possible+pixel+values+that+could+plausibly+complete+occluded+image+regions.+State-of-the+art+supervised+learning+methods+are+typically+optimized+to+make+a+single+test-time+prediction+for+each+query%2C+failing+to+find+other+modes+in+the+output+space.+Existing+methods+that+allow+for+sampling+often+sacrifice+speed+or+accuracy.+We+introduce+a+simple+method+for+training+a+neural+network%2C+which+enables+diverse+structured+predictions+to+be+made+for+each+test-time+query.+For+a+single+input%2C+we+learn+to+predict+a+range+of+possible+answers.+We+compare+favorably+to+methods+that+seek+diversity+through+an+ensemble+of+networks.+Such+stochastic+multiple+choice+learning+faces+mode+collapse%2C+where+one+or+more+ensemble+members+fail+to+receive+any+training+signal.+Our+best+performing+solution+can+be+deployed+for+various+tasks%2C+and+just+involves+small+modifications+to+the+existing+single-mode+architecture%2C+loss+function%2C+and+training+regime.+We+demonstrate+that+our+method+results+in+quantitative+improvements+across+three+challenging+tasks%3A+2D+image+completion%2C+3D+volume+estimation%2C+and+flow+prediction.&rft.subject=Training%2C+Task+analysis%2C+Aerospace+electronics%2C+Three-dimensional+displays%2C+Supervised+learning%2C+Two+dimensional+displays%2C+Training+data&rft.publisher=IEEE&rft.date=2018-12-17&rft.type=Proceedings+paper&rft.publisher=31st+IEEE%2FCVF+Conference+on+Computer+Vision+and+Pattern+Recognition+(CVPR)&rft.language=eng&rft.source=+++++In%3A++Proceedings+of+the+2018+IEEE%2FCVF+Conference+on+Computer+Vision+and+Pattern+Recognition.++(pp.+pp.+5598-5607).++IEEE%3A+Salt+Late+City%2C+UT%2C+USA.+(2018)+++++&rft.format=text&rft.identifier=https%3A%2F%2Fdiscovery.ucl.ac.uk%2Fid%2Feprint%2F10074121%2F1%2Fcvpr18_diversenet.pdf&rft.identifier=https%3A%2F%2Fdiscovery.ucl.ac.uk%2Fid%2Feprint%2F10074121%2F&rft.rights=open