?url_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Adc&rft.title=Customizing+360-Degree+Panoramas+Through+Text-to-Image+Diffusion+Models&rft.creator=Wang%2C+Hai&rft.creator=Xiang%2C+Xiaoyu&rft.creator=Fan%2C+Yuchen&rft.creator=Xue%2C+Jinghao&rft.description=Personalized+text-to-image+(T2I)+synthesis+based+on+diffusion+models+has+attracted+significant+attention+in+recent+research.+However%2C+existing+methods+primarily+concentrate+on+customizing+subjects+or+styles%2C+neglecting+the+exploration+of+global+geometry.+In+this+study%2C+we+propose+an+approach+that+focuses+on+the+customization+of+360-degree+panoramas%2C+which+inherently+possess+global+geometric+properties%2C+using+a+T2I+diffusion+model.+To+achieve+this%2C+we+curate+a+paired+image-text+dataset+specifically+designed+for+the+task+and+subsequently+employ+it+to+fine-tune+a+pre-trained+T2I+diffusion+model+with+LoRA.+Nevertheless%2C+the+fine-tuned+model+alone+does+not+ensure+the+continuity+between+the+leftmost+and+rightmost+sides+of+the+synthesized+images%2C+a+crucial+characteristic+of+360-degree+panoramas.+To+address+this+issue%2C+we+propose+a+method+called+StitchDiffusion.+Specifically%2C+we+perform+pre-denoising+operations+twice+at+each+time+step+of+the+denoising+process+on+the+stitch+block+consisting+of+the+leftmost+and+rightmost+image+regions.+Furthermore%2C+a+global+cropping+is+adopted+to+synthesize+seamless+360-degree+panoramas.+Experimental+results+demonstrate+the+effectiveness+of+our+customized+model+combined+with+the+proposed+StitchDiffusion+in+generating+high-quality+360-degree+panoramic+images.+Moreover%2C+our+customized+model+exhibits+exceptional+generalization+ability+in+producing+scenes+unseen+in+the+fine-tuning+dataset.+Code+is+available+at+https%3A%2F%2Fgithub.com%2Flittlewhitesea%2FStitchDiffusion.&rft.publisher=IEEE&rft.date=2024-04-09&rft.type=Proceedings+paper&rft.language=eng&rft.source=+++++In%3A++2024+IEEE%2FCVF+Winter+Conference+on+Applications+of+Computer+Vision+(WACV).++(pp.+pp.+4933-4943).++IEEE%3A+Waikoloa%2C+HI%2C+USA.+(2024)+++++&rft.format=text&rft.identifier=https%3A%2F%2Fdiscovery.ucl.ac.uk%2Fid%2Feprint%2F10184511%2F1%2FWang_Customizing_360-Degree_Panoramas_Through_Text-to-Image_Diffusion_Models_WACV_2024_paper.pdf&rft.identifier=https%3A%2F%2Fdiscovery.ucl.ac.uk%2Fid%2Feprint%2F10184511%2F&rft.rights=open