2022/9/14 GPU UNITE 2022 / Day3ï¼CGç 究 ã»ãã·ã§ã³ (https://gdep-sol.co.jp/gpu-event/gpu-unite-2022/day3/) è¬æ¼è³æ
Preferred Networksã¯4æ26æ¥ãé¡ãªãã³ã«ä¸å身ã®ã¤ã©ã¹ããèªåçæã»ç·¨éã§ãããã£ã©ã¯ã¿ã¼çæãã©ãããã©ã¼ã ãCrypkoãï¼ã¯ãªãã³ï¼ããåæ¥ããã¦ã§ããµã¼ãã¹ã¨ãã¦æä¾ãéå§ããã Crypkoã¯ã深層å¦ç¿ãå¿ç¨ããç»åçæã¢ãã«ã®1ã¤ã§ãããæµå¯¾ççæãããã¯ã¼ã¯ãï¼generative adversarial networksãGANï¼ã使ã£ãé¡ã®ã¤ã©ã¹ãçæã¢ãã«ã¨ãã¦ã2018å¹´ã«éçºãéå§ã6ä¸ä»£ç®ã¨ãªãä»åã®çæã¢ãã«ã§ã¯ãé¡ã ãã§ãªã人éã®ã¤ã©ã¹ãã¬ã¼ã¿ã¼ãæç»ãããããªãã£ã©ã¯ã¿ã¼ã®ä¸å身ã®ã¤ã©ã¹ããçæãããã¨ãå¯è½ã¨ãªã£ã¦ããã ããã«ãçæãããã£ã©ã¯ã¿ã¼ãèåãããã¨ã«ããããããã®ç¹å¾´ãç¶æ¿ããæ°ãããã£ã©ã¯ã¿ã¼ãçæããã髪ãç®ã®è²ã表æ ãªã©ã®å±æ§ãã¦ã¼ã¶ã¼ã®å¥½ã¿ã§å¤æ´ãããã¨ãå¯è½ã¨ãªã£ã¦ãããã¦ã¼ã¶ã¼ã¯ãæå®ã®ã¢ããªå ãã¤ã³ãï¼Cr
Innovative Techï¼ ãã®ã³ã¼ãã¼ã§ã¯ããã¯ããã¸ã¼ã®ææ°ç 究ãç´¹ä»ããWebã¡ãã£ã¢ãSeamlessãã主宰ããå±±ä¸è£æ¯ æ°ãå·çãæ°è¦æ§ã®é«ãç§å¦è«æãå±±ä¸æ°ãããã¯ã¢ãããã解説ããã NVIDIAã¨ç±³Cornell Universityã®ç 究ãã¼ã ãéçºãããGANcraft: Unsupervised 3D Neural Rendering of Minecraft Worldsãã¯ããã¤ã³ã¯ã©ãããªã©ã®ãã¯ã»ã«ãã¼ã¹ã®å¤§è¦æ¨¡ãªä¸çãåå®çãªç»åã«å¤æããææ³ã ãåãç ãèæ¨ãéªãæµ·ã空ãªã©ããªã¢ã«ã«è¡¨ç¾ããè¦ãè¦ç¹ã移åãã¦ãä¸è²«ããç»åãæä¾ãããã²ã¼ã ãã¢ãã¡ã¼ã·ã§ã³ã®èæ¯ããã¼ãã£ã«ç©ºéã®ä½æã«ãå½¹ç«ã¤å¯è½æ§ãç§ãã¦ããã ãã¤ã³ã¯ã©ããã¯ãè¦åçãªã°ãªããä¸ã«ãããã¯ãç½®ãã ãã§3次å 空éãæ§ç¯ã§ãããåã©ãã§ãå¯è½ãªæ軽ããé åã ããå3Dãããã¯ã¯å¤§ãã
Innovative Techï¼ ãã®ã³ã¼ãã¼ã§ã¯ããã¯ããã¸ã¼ã®ææ°ç 究ãç´¹ä»ããWebã¡ãã£ã¢ãSeamlessãã主宰ããå±±ä¸è£æ¯ æ°ãå·çãæ°è¦æ§ã®é«ãç§å¦è«æãå±±ä¸æ°ãããã¯ã¢ãããã解説ããã ç±³ã¤ãªãã¤å¤§å¦ã¢ã¼ããã»ã·ã£ã³ãã¼ã³æ ¡ã®ç 究ãã¼ã ãéçºãããJoJoGAN: One Shot Face Stylizationãã¯ãå ¥åãã1æã®é¡ç»åã漫ç»ãã¸ã§ã¸ã§ã®å¥å¦ãªåéºãã®ãã£ã©ã¯ã¿ã¼é¢¨ã«å¤æããæ©æ¢°å¦ç¿ãã¬ã¼ã ã¯ã¼ã¯ã ãã空æ¡æ¿å¤ªéããªã©ãä½ä¸ã®ãã£ã©ã¯ã¿ã¼ã¹ã¿ã¤ã«ã«ä¼¼ããé¡ã«å¤æããããã®ãã¬ã¼ã ã¯ã¼ã¯ã¯ã大éã®ãã¢ç»åãã¼ã¿ã»ããã使ããã1æã®é¡ç»åããå¦ç¿ã§ããç¹æ§ãæã¤ã ç»åãã¹ã¿ã¤ã«å¤æããã¢ãã«ã¯ã大éã®ãã¢ç»åãã¼ã¿ã»ãããå¿ è¦ã¨ãããããã®ãã¼ã¿ãæºåããã«ã¯ã³ã¹ããããããæ°æã®ç»åããã¹ã¿ã¤ã«å¤æããç 究ãå ±åããã¦ãããã詳細ãªã¹ã¿ã¤ã«ãå¤æ§æ§ã
ã¹ããã¯ãã©ããµã¼ãã¹ãææããã¤ã¡ã¼ã¸ããï¼æå¹å¸ï¼ã¯6æ1æ¥ãAIãæ´»ç¨ãã¦å®å¨ããªãã¢ãã«ã®ç»åãä½æãããµã¼ãã¹ãINAI MODELãã«ãç·æ§ã¢ãã«ã追å ãããããã¾ã§å¥³æ§ã®ç»åã®ã¿æä¾ãã¦ããããAIã®ã¢ãããã¼ããéããç·æ§ã®å ´åã§ãã¯ãªãªãã£ã¼ãç¶æã§ããã¨å¤æãããã¨ãã販売ã決ããã¨ãããå¾æ¥ã¨åãããWebåºåããã¹ã¿ã¼ãªã©ã§ã®å©ç¨ãè¦è¾¼ãã INAI MODELã§ã¯ãå®å¨ãã人ç©ã®åçãå©ç¨è¨±è«¾ãå¾ãä¸ã§æ®å½±ãAIã§é¡é¨åãå å·¥ãã¦æ¶ç©ºã®ã¢ãã«ç»åãçæãããéçºã«ã¯AIãã³ãã£ã¼ã®ãã¼ã¿ã°ãªããï¼äº¬é½å¸ï¼ãååãç»åãçæããAIã¨ç»åãAIã«ãããã®ãè©ä¾¡ããAIãæ¦ããããã¨ã§ç²¾åº¦ãåä¸ãããæè¡ãGANãï¼æµå¯¾ççæãããã¯ã¼ã¯ï¼ãæ´»ç¨ãã¦ããã ã¢ãã«ãå®å¨ããªããããä¸åº¦è³¼å ¥ããã°å¥ç´æéãæ°ã«ããæ°¸ä¹ ã«ç»åã使ããä»ãã¢ãã«ãã¹ãã£ã³ãã«ãèµ·ãããªã¹ã¯
The Morning After: Should you upgrade to an iPhone 16?
ã赤ã¡ããACãã®ã¹ããå°ç¨ã¢ããªããªãªã¼ã¹ï¼ 2æã®åçãã赤ã¡ããã®é¡ãäºæ¸¬ï¼ 赤ã¡ããã®é¡ã¯ãçå¾ãæ人ã¾ã§ã®ï¼æ®µéäºæ¸¬ãå¯è½(ä¸é¨æææ©è½ãã) StyleGANã«ããã親ã®é¡ã®ç¹å¾´ããã£ãé«è§£å度ã®èµ¤ã¡ããç»åãçæãã¾ã 赤ã¡ããACã§ã¯StyleGANã¨ããAIæè¡ãç¨ãã¦ã赤ã¡ããã®ç»åãçæãã¦ãã¾ããStyleGANã¯GANï¼æµå¯¾ççæãããã¯ã¼ã¯ï¼ã¨ãããã£ã¼ãã©ã¼ãã³ã°ã®ã¢ãã«ãçºå±ããããã®ã§ãï¼æã®ç»åã®ç¹å¾´ãå¦ç¿ããåæãã¦é«è§£å度ã®æ°ããç»åãçæãããã¨ãã§ãã¾ãã 赤ã¡ããACã§ã¯ã親ã¨ãªãï¼æã®é¡åçããé¡ã®ç¹å¾´ãæ½åºããã©ã³ãã ã«é¸ã°ãã赤ã¡ããã®åçã«ãããã®ç¹å¾´ãåæãããã¨ã§ã親ã®ç¹å¾´ãæã£ãé«è§£å度ã®èµ¤ã¡ããã®é¡åçãçæãã¾ãã ã¾ãã親ã®é¡ã®ç¹å¾´ãé«éã«æ½åºï¼ã¨ã³ã³ã¼ãï¼ããããã«ãpixel2Style2pixelã¨ããã¨ã³ã³ã¼ãæè¡ã
ç»åç´ æãµã¤ããéå¶ããACã¯ã¼ã¯ã¹ã¯3æ15æ¥ã両親ã®é¡åçãã赤ã¡ããã®é¡ãAIã§äºæ¸¬ãããµã¼ãã¹ã赤ã¡ããACããå ¬éãããç¡æã§å©ç¨ã§ããã 2人ã®é¡åçãåãµã¤ãã¸ã¢ãããã¼ãããã¨ã2人ããçã¾ãã赤ã¡ããã®é¡ç»åãAIã§åæãããåæç»åã¯ç¡æã§ãã¦ã³ãã¼ãå¯è½ã§ãã¢ãããã¼ãããé¡åçã24æéã§å®å ¨ã«æ¶å»ããã¨ãã¦ããã 赤ã¡ããã®é¡ç»åã®åæã«ã¯ã2人ã®é¡ãçµã¿åããã¦é«è§£åãªåæç»åãä½ããã£ã¼ãã©ã¼ãã³ã°æè¡ã®ä¸ã¤ã§ãããç±³NVIDIAã®ãStyleGANããå©ç¨ãã¦ããã¨ããã親ã®é¡ã®ç¹å¾´æ½åºã«ã¯ãpixel2Style2pixelãã¨ããStyleGANåãã®è¨ç®ææ³ã使ããã¨ã§ãé常ã¯1åã»ã©ãããã¨ã³ã³ã¼ãå¦çãæ°ç§ã«ã¾ã§ç縮ããã¨ãã¦ããã é¢é£è¨äº âå®å¨ããªããªã¢ã«ãªé¡âãèªå¨ã«ç·¨éã§ãããStyleRigããStyleGANã§çæããé¡ã®åãã表æ ã
AIæè¡ãç¨ãã¦æ¶ç©ºã®ç»åãæ åãåæãããã£ã¼ããã§ã¤ã¯ã¯ã人ã®ç®ã«ã¯çå½ãè¦åãããã¨ãé常ã«é£ãã段éã¾ã§çºå±ãã¦ãããå®å¨ããªã人ç©ã®é¡åçãç°¡åã«ä½æã§ãããã¨ããSNSãªã©ã§ã大ããªåé¡ã¨ãªã£ã¦ãã¾ãããããªãã£ã¼ããã§ã¤ã¯ã®é¡åçããç³ã«æ ã£ãå ã®åå°ããè¦æãæè¡ãããã¥ã¼ã¨ã¼ã¯å·ç«å¤§å¦ãããã¡ãã¼æ ¡ã®ç 究ãã¼ã ãéçºãã¾ããã EXPOSING GAN-GENERATED FACES USING INCONSISTENT CORNEAL SPECULAR HIGHLIGHTS (PDFãã¡ã¤ã«)https://arxiv.org/pdf/2009.11924.pdf New AI tool detects Deepfakes by analyzing light reflections in the eyes https://thenextweb.com/neural/2
About Service ãµã¼ãã¹ã«ã¤ã㦠AIçææ¶ç©ºã¢ãã«Â  ã¨ã¯ ãINAI MODELãã¯äººç©åçã®é¡é¨åããAIãçæããå®å¨ããªãé¡ã¨å ¥ãæ¿ããåæåçã¢ãã«ã§ãã ããã¾ã§ã¢ãã«æ®å½±ã«ããã¦ããè²»ç¨ãæéãªã©ã®ã³ã¹ãåæ¸ã«å ããã¢ãã«ã®å¥ç´æéã®ç®¡çã ã¹ãã£ã³ãã«ãªã¹ã¯ãåé¿ããæ§ã ãªã¡ãªããã§å®å¿ããå±ããã¾ãã ãINAI MODELãã¯äººç©åçã®é¡é¨åããAIãçæããå®å¨ããªãé¡ã¨å ¥ãæ¿ããåæåçã¢ãã«ã§ãã ããã¾ã§ã¢ãã«æ®å½±ã«ããã¦ããè²»ç¨ãæéãªã©ã®ã³ã¹ãåæ¸ã«å ããã¢ãã«ã®å¥ç´æéã®ç®¡çãã¹ãã£ã³ãã«ãªã¹ã¯ãåé¿ããæ§ã ãªã¡ãªããã§å®å¿ããå±ããã¾ãã
åçç´ æã®è²©å£²ãªã©ãææããã¤ã¡ã¼ã¸ããã¯6æ8æ¥ãAIãå®å¨ããªãã¢ãã«ã®ç»åãçæãããµã¼ãã¹ãINAI MODELããå§ãããç»åãè³¼å ¥ããä¼æ¥ã¯ãæ¶ç©ºã®ã¢ãã«ãWebåºåããã¹ã¿ã¼ãªã©ã«èµ·ç¨ã§ãããã¢ãã«ã¨ã®å¥ç´æéãæ°ã«ããæ°¸ä¹ ã«ç»åã使ããä»ãã¢ãã«ãã¹ãã£ã³ãã«ãèµ·ãããªã¹ã¯ããªããå·®ãæ¿ããçºçããªãã®ãç¹å¾´ã å½åã¯è¥ã女æ§ã®ç»åã®ã¿çæã»è²©å£²ããä»å¾ã¯ç·æ§ãã·ãã¢ã«ã対å¿ããäºå®ãåãã¢ãã«ãç°ãªããã¼ãºãã¨ã£ã¦ããç»åããç°ãªãã¢ãã«ãåããã¼ãºãã¨ã£ã¦ããç»åãªã©ãçæã§ãããèå人ã«ä¼¼ããã¢ãã«ã¯ä½æããªãã ãµã¼ãã¹éçºã«ã¯AIãã³ãã£ã¼ã®ãã¼ã¿ã°ãªãããååãç»åãçæããAIã¨ç»åãè©ä¾¡ããAIãæµå¯¾ããã精度ãåä¸ãããæè¡ãGANãï¼æµå¯¾ççæãããã¯ã¼ã¯ï¼ãå©ç¨ããã ã¢ãã«ç»åãçæããéã¯ãã¤ã¡ã¼ã¸ãããå©ç¨è¨±è«¾ãå¾ãä¸ã§å®å¨ãã人ç©ã®åçãæ®å½±ã
ãã§ã³å·¥ç§å¤§å¦ã¨Snapã«ããç 究ãã¼ã ã¯ãããç»é¢¨ã¹ã¿ã¤ã«ãç»åã«è»¢éãã深層å¦ç¿ãç¨ããã¹ã¿ã¤ã«å¤ææè¡ãInteractive Video Stylization Using Few-Shot Patch-Based Trainingããçºè¡¨ãããã¹ã¿ã¤ã«å¤æã¨ã¯ãææ¸ã風ã®ã¹ã¿ã¤ã«ãç»åãåç»ã«è»¢éããã³ã³ãã³ãã®ååã¯æ®ãã¤ã¤ã¹ã¿ã¤ã«ã ããå¤æããæè¡ã ã Interactive Video Stylization Using Few-Shot Patch-Based Trainingã®åºåçµæ ãããã£ãã¹ã¿ã¤ã«å¤æåéã§ã¯ãç»åã¸è»¢éããæè¡ã¨ãã¦2016å¹´ã«ãImage Style Transfer Using Convolutional Neural Networksã[Gatys et al. 2016]ã§çºè¡¨ããã¦ãããCNNãç¨ããVGGã使ç¨ããã³ã³ãã³ãã¨ã¹ã¿ã¤
å°è±¡ã®æä½ãè¡ããã¨è¨ã£ãâå®æ§çãªâç»åã®æä½ã¯AIã«ã¯ããã¾ã§é£ããã¨ãããã¦ãã¾ããã ä¾ãã°ãâè¨æ¶ãããããâç»åã¨ã¯ã©ã®ãããªãã®ã§ããããï¼è¨æ¶ã®ããããã¯ããããã«ä¼´ããã®å ·ä½çãªå®ç¾©ãããã¾ããã ä»åç´¹ä»ããæ°ããªã¢ã¤ãã¢ã§ã¯ãâè¨æ¶ãããããâã¨ããç»åã®æä½ãã©ã®ãããªãã®ãå¦ç¿ãæé©åãããã¨ã§ãä»»æã®ç»åã人éã«ã¨ã£ã¦è¨æ¶ãããããããã«å¤åããã¾ãã è¨æ¶ã®ããããç»åããâç©ä½ã大ããã»ç»åãæããã»å½¢ç¶ã丸ãâããã°è¯ããã¨ãã風ã«å ·ä½çãªæä½ã«è½ã¨ãè¾¼ã¿ãå¾åãå¦ç¿ããããã¨ã§çæãå¯è½ã«ãã¾ãã ããã«ãå¦ç¿ãããã©ã¡ã¼ã¿ã«âéã¿ä»ãâãè¡ããã¨ãä»åã¯è©¦ã¿ã¾ããããã«ãããâè¨æ¶ã®ããããâã¨ããæ¹åæ§ãå ã«ãã©ã®ç¨åº¦è¨æ¶ãããããã調æ´ãããã¨ãå¯è½ã§ãã ä¾ãã°ãä¸ã®ãã¼ãºãã¼ã¬ã¼ãèãã¦ã¿ã¦ãã ããããã®ã¢ãã«ã§ã¯ãå³ã«ç§»åããã«ã¤ãã¦ãããè¨
GANPaint Studio is a demonstration how, with the help of two neural networks (GAN and Encoder). It's easy to start drawing: Select an image Select if you want to draw (paintbrush) or delete (eraser) Select a semantic paintbrush (tree,grass,..) Enjoy painting! Beware -- the model has a character of its own! Come back often, as we will add new and better models over time. For more information about
ã©ã³ãã³ã°
ã©ã³ãã³ã°
ã©ã³ãã³ã°
ãªãªã¼ã¹ãé害æ å ±ãªã©ã®ãµã¼ãã¹ã®ãç¥ãã
ææ°ã®äººæ°ã¨ã³ããªã¼ã®é ä¿¡
å¦çãå®è¡ä¸ã§ã
j次ã®ããã¯ãã¼ã¯
kåã®ããã¯ãã¼ã¯
lãã¨ã§èªã
eã³ã¡ã³ãä¸è¦§ãéã
oãã¼ã¸ãéã
{{#tags}}- {{label}}
{{/tags}}