[B! GAN] tsu-neraã®ãƒ–ãƒƒã‚¯ãƒžãƒ¼ã‚¯

tsu-nera id:tsu-nera

GANã«é–¢ã™ã‚‹tsu-neraã®ãƒ–ãƒƒã‚¯ãƒžãƒ¼ã‚¯ (14)

${{author_name}}$

{{{comment_expanded}}}

{{label}}

{{#is_bookmark}}ãƒªã‚¹ãƒˆ{{/is_bookmark}}{{^is_bookmark}}ãƒªãƒ³ã‚¯{{/is_bookmark}}

${{author_name}}$
{{author_name}}{{created}}
{{ #comment }}{{ comment }}{{ /comment }}
- {{ label }}

${{author_name}}$

{{{comment_expanded}}}

{{label}}

{{#is_bookmark}}ãƒªã‚¹ãƒˆ{{/is_bookmark}}{{^is_bookmark}}ãƒªãƒ³ã‚¯{{/is_bookmark}}

infoGANã‚’å®Ÿè£…ã—ãŸ(ã„) - akmtnè¨˜éŒ²
infoGANã®è«–æ–‡ã‚’èªã¿ï¼ŒMNISTç”¨ã®å®Ÿè£…ã‚’PyTorchã§è¡Œã£ãŸè¨˜éŒ²ã§ã™ï¼Ž è«–æ–‡ã¯2016å¹´6æœˆã«å‡ºã¦ã„ã‚‹ã®ã§1å¹´ã»ã©å‰ã®ã‚‚ã®ï¼Ž [1606.03657] InfoGAN: Interpretable Representation Learning by Information Maximizing Generative Adversarial Nets ã‚¿ã‚¤ãƒˆãƒ«ã‚’æ—¥æœ¬èªžè¨³ã™ã‚‹ã¨ï¼Œã€Œæƒ…å ±é‡ã‚’æœ€å¤§åŒ–ã™ã‚‹æ•µå¯¾çš„ç”Ÿæˆãƒãƒƒãƒˆãƒ¯ãƒ¼ã‚¯ã«ã‚ˆã‚‹è§£é‡ˆå¯èƒ½ãªè¡¨ç¾ã®å¦ç¿’ã€ã§ã—ã‚‡ã†ã‹ï¼Ž infoGANã®ç‰¹å¾´ ãƒ»GAN ãƒ»æ˜Žå¿«ãªè¡¨ç¾ã‚’ç²å¾—ã™ã‚‹ ãƒ»æ•™å¸«ãªã—å¦ç¿’ ãƒ»ç›¸äº’æƒ…å ±é‡ã®æœ€å¤§åŒ–ï¼ˆéš ã‚Œç¬¦å·ã¨ç”Ÿæˆç”»åƒé–“ã®ï¼‰ ãƒ»DCGANã®ä¸€éƒ¨ã‚’å¤‰ãˆã‚‹ã ã‘ã§ä½œã‚Œï¼Œè¨“ç·´ã‚‚ç°¡å˜ ç›¸äº’æƒ…å ±é‡ã¯ï¼Œï¼’ã¤ã®ç¢ºçŽ‡å¤‰æ•°ã®ã†ã¡ï¼‘ã¤ãŒã‚ã‹ã£ãŸæ™‚ã«ï¼Œã‚‚ã†ï¼‘ã¤ã®ç¢ºçŽ‡å¤‰æ•°ã«é–¢ã—ã¦ã©ã‚Œã»ã©æŽ¨æ¸¬ã§ãã‚‹ã‹ã‚’ç¤ºã™ã®ã§ï¼Œï¼’ã¤ãŒç‹¬ç«‹ãªæ™‚ã«ã¯ç›¸äº’æƒ…å ±é‡ã¯ï¼ã«ãªã‚Šã¾ã™ï¼Ž
tsu-nera 2017/08/06
GAN

æƒ…å ±ç†è«–
ãƒªãƒ³ã‚¯
GAN(Generative Adversarial Networks)ã‚’å¦ç¿’ã•ã›ã‚‹éš›ã®14ã®ãƒ†ã‚¯ãƒ‹ãƒƒã‚¯ - Qiita
Deleted articles cannot be recovered. Draft of this article would be also deleted. Are you sure you want to delete this article? â€»ã“ã®è¨˜äº‹ã¯"How to Train a GAN?" at NIPS2016ã‚’ã€ã“ã“ã‚’å‚è€ƒã«ã—ã¦ã€ç§ãªã‚Šã«è§£é‡ˆã—ã¦ã€é‡è¦ãã†ãªéƒ¨åˆ†ã‚’å¼•ç”¨ã—ã¦ç¿»è¨³ã—ãŸã‚‚ã®ã§ã™ã€‚å½¹ã«ç«‹ã¤ã“ã¨ãŒå¤šã‹ã£ãŸã®ã§å…±æœ‰è‡´ã—ã¾ã™ã€‚ â€»GANã®èª¬æ˜Žã¯çœç•¥ã—ã¾ã™ã€‚G=Generatorã€D=Discriminatorã¨ã—ã¦èª¬æ˜Žã—ã¾ã™ã€‚ â€»ãã‚Œãžã‚Œã®ãƒ†ã‚¯ãƒ‹ãƒƒã‚¯ã«æ ¹æ‹ ã¯ã‚ã¾ã‚Šæ›¸ã„ã¦ãªã„ã§ã™ãŒã€ã»ã¨ã‚“ã©ãŒè«–æ–‡ã«æ›¸ã„ã¦ã‚ã£ãŸã€Œã“ã†ã™ã‚Œã°ã†ã¾ãã„ã£ãŸã€ã¨ã„ã†ã‚‚ã®ã§ã™ã€‚GANã®å¦ç¿’ãŒã†ã¾ãã„ã‹ãªã„ã¨ãã«è©¦ã—ã¦ã¿ã‚‹ã¨è‰¯ã„ã¨æ€ã„ã¾ã™ã€‚ 1. å…¥åŠ›ã‚’æ£è¦åŒ– (Normalize the
tsu-nera 2017/08/06
GAN
ãƒªãƒ³ã‚¯
ã‚·ãƒ³ãƒ—ãƒ«ãªGANã®å®Ÿè£… - akmtnè¨˜éŒ²
GANã®è«–æ–‡ã‚’èªã‚“ã ã¤ã„ã§ã«MNISTç”¨ã«å®Ÿè£…ã—ã¾ã—ãŸï¼Ž ãƒãƒƒãƒˆãƒ¯ãƒ¼ã‚¯ã«ã¯ï¼ŒConvolutionã‚’ä½¿ã‚ãšï¼Œå¤šå±¤ãƒ‹ãƒ¥ãƒ¼ãƒ©ãƒ«ãƒãƒƒãƒˆã§ã™ï¼Žå±¤ã®æ•°ï¼Œæ´»æ€§åŒ–é–¢æ•°ã‚„ãƒãƒƒãƒæ£è¦åŒ–ãªã©ã®è¨å®šã¯ã¦ãã¨ã†ã§ã™ï¼Žãªã®ã§GANã‚‚ã©ãã§ã™ï¼Ž ãŒã‚“ã‚‚ã©ã pytorchã§å®Ÿè£…ã—ãŸã‹ã£ãŸã®ã§ï¼Œæœ¬å®¶ã®DCGANã®ã‚³ãƒ¼ãƒ‰ã‚’å‚è€ƒã«ã—ãªãŒã‚‰æ›¸ãã¾ã—ãŸï¼Ž github.com ã¾ãŸï¼Œã“ã¡ã‚‰ã®æ–¹ã®å®Ÿè£…ã‚‚å‚è€ƒã«ã—ã¾ã—ãŸï¼Ž(keras) yusuke-ujitoko.hatena blog.com çµæžœã¨ã—ã¦ï¼Œä»¥ä¸‹ã®ã‚ˆã†ãªå‡ºåŠ›ã‚’å¾—ã¾ã—ãŸï¼Ž ãã‚Œãªã‚Šã«æ•°å—ã«ã¯è¦‹ãˆã¾ã™ãŒï¼Œï¼‘ãŒå¤šã„ä¸€æ–¹ï¼Œï¼’ãŒå°‘ãªã„ã¨ã„ã†åã‚ŠãŒã‚ã‚Šã¾ã—ãŸï¼Ž ä¸€å¿œï¼Œã‚³ãƒ¼ãƒ‰ã¯ã“ã“ã«ã‚ã‚Šã¾ã™ï¼Ž
tsu-nera 2017/08/05
GAN

PyTorch
ãƒªãƒ³ã‚¯
Generative Adversarial Networks (GANs) in 50 lines of code (PyTorch)
In 2014, Ian Goodfellow and his colleagues at the University of Montreal published a stunning paper introducing the world to GANs, or generative adversarial networks. Through an innovative combination of computational graphs and game theory they showed that, given enough modeling power, two models fighting against each other would be able to co-train through plain old backpropagation. The models p
tsu-nera 2017/08/05
PyTorch

GAN
ãƒªãƒ³ã‚¯
Wasserstein GAN Tfug2017 07-12
1. The document discusses Wasserstein GAN (WGAN), a type of generative adversarial network (GAN) that uses the Wasserstein distance rather than Jensen-Shannon divergence. WGAN has improved stability during training over traditional GANs. 2. WGAN trains the discriminator/critic to estimate the Wasserstein distance between real and generated distributions rather than classify samples. The gradient o
tsu-nera 2017/08/05
GAN

WGAN
ãƒªãƒ³ã‚¯
ã§ãã‚‹ã ã‘ä¸å¯§ã«GANã¨DCGANã‚’ç†è§£ã™ã‚‹ - åˆç¡äºŒæ™‚å››ååˆ†
ç›®çš„ Chainerã®æ‰±ã„ã«æ…£ã‚Œã¦ããŸã®ã§ã€ãƒ‹ãƒ¥ãƒ¼ãƒ©ãƒ«ãƒãƒƒãƒˆãƒ¯ãƒ¼ã‚¯ã‚’ä½¿ã£ãŸç”»åƒç”Ÿæˆã«æ‰‹ã‚’å‡ºã—ã¦ã¿ãŸã„ ã„ã‚ã„ã‚ãªæ‰‹æ³•ãŒææ¡ˆã•ã‚Œã¦ã„ã‚‹ãŒã€ã¾ãšã¯ä»Šå¹´å§‹ã‚ã«è©±é¡Œã«ãªã£ãŸDCGANã‚’å®Ÿéš›ã«è©¦ã—ã¦ã¿ã‚‹ãŸã„ ãã®ãŸã‚ã«ã€ DCGANã‚’ã§ãã‚‹ã ã‘ä¸å¯§ã«ç†è§£ã™ã‚‹ã“ã¨ãŒã“ã®ã‚¨ãƒ³ãƒˆãƒªã®ç›®çš„ å°†æ¥GAN / DCGANã‚’è§¦ã‚‹äººã®åŠ©ã‘ã«ãªã£ãŸã‚Šã€ç†è§£é–“é•ã£ã¦ã‚‹ã¨ã“ã‚ã«ãƒ„ãƒƒã‚³ãƒŸãŒã‚ã‚‹ã¨å¬‰ã—ã„ æœ¬ã‚¨ãƒ³ãƒˆãƒªã®æ§‹æˆ DCGANã®å‰æã¨ãªã£ã¦ã„ã‚‹GANã®è«–æ–‡ã®è¦ç‚¹ã‚’ã¾ã¨ã‚ã‚‹ DCGANã§GANã¨ã®å·®åˆ†ã¨ã—ã¦ææ¡ˆã•ã‚Œã¦ã„ã‚‹è¦ç‚¹ã‚’ã¾ã¨ã‚ã‚‹ DCGANã®mattyaã•ã‚“ã®å®Ÿè£…ã‚’èªã¿é€šã—ã¦è©³ç´°ã‚’ç†è§£ã™ã‚‹ 1. GANã«ã¤ã„ã¦ GANã¯ã€ã‚µãƒ³ãƒ—ãƒ«ç¾¤ã¨ä¼¼ãŸã‚ˆã†ãªæ€§è³ªã‚’æŒã¤å‡ºåŠ›ã‚’ç”Ÿæˆã™ã‚‹ãŸã‚ã®ãƒ•ãƒ¬ãƒ¼ãƒ ãƒ¯ãƒ¼ã‚¯ 2014å¹´ã«Ian J. Goodfellowã‚‰ã«ã‚ˆã£ã¦ææ¡ˆã•ã‚ŒãŸ è«–æ–‡: Generative Adversarial Net
tsu-nera 2017/08/05
DCGAN

GAN
ãƒªãƒ³ã‚¯
ã¯ã˜ã‚ã¦ã®GAN
ä»Šå›žã¯GANï¼ˆGenerative Adversarial Networkï¼‰ã‚’è§£èª¬ã—ã¦ã„ãã¾ã™ã€‚ GANã¯â€œDeep Learningâ€ã¨ã„ã†æœ¬ã®è‘—è€…ã§ã‚‚ã‚ã‚‹Ian GoodfellowãŒè€ƒæ¡ˆã—ãŸãƒ¢ãƒ‡ãƒ«ã§ã™ã€‚NIPS 2016ã§ã‚‚GANã®ãƒãƒ¥ãƒ¼ãƒˆãƒªã‚¢ãƒ«ãŒè¡Œã‚ã‚Œã‚‹ãªã©éžå¸¸ã«æ³¨ç›®ã‚’é›†ã‚ã¦ã„ã‚‹åˆ†é‡Žã§ã€æ¬¡ã€…ã«è«–æ–‡ãŒå‡ºã¦ãã¦ã„ã¾ã™ã€‚ ã¾ãŸã€Quoraã®ã‚»ãƒƒã‚·ãƒ§ãƒ³ã§Yann LeCunãŒã€ã“ã®10å¹´ã®æ©Ÿæ¢°å¦ç¿’ã§æœ€ã‚‚é¢ç™½ã„ã‚¢ã‚¤ãƒ‡ã‚£ã‚¢ã¨è¿°ã¹ã¦ã„ãŸã‚Šã‚‚ã—ã¾ã™ã€‚ â€œThe most interesting idea in the last 10 years in ML, in my opinion.â€ â€“Yann LeCun GANã¯èžã„ãŸã“ã¨ã¯ã‚ã‚‹ã‘ã‚Œã©ã‚ã¾ã‚Šè¿½ãˆã¦ãªã„ã¨ã„ã†äººå‘ã‘ã«åŸºç¤Žã‹ã‚‰è§£èª¬ã—ã¦ã„ããŸã„ã¨æ€ã„ã¾ã™ã€‚ãã‚Œã§ã¯é †ã«è¦‹ã¦ã„ãã¾ã—ã‚‡ã†ã€‚ ç›®æ¬¡ åŸºç¤Žç†è«– DCGAN å®Ÿè£… è«–æ–‡ç´¹ä»‹ ã¾ã¨ã‚ åŸºç¤Žç†
tsu-nera 2017/08/05
GAN

DCGAN
ãƒªãƒ³ã‚¯
Deep Convolutional GANs(DCGAN)ã‚’kerasã§å®Ÿè£…ã—ã¦ã€ã„ã‚‰ã™ã¨ã‚„ç”»åƒã‚’ç”Ÿæˆã™ã‚‹ - ç·‘èŒ¶æ€è€ƒãƒ–ãƒã‚°
å‰å›žï¼ŒGANã‚’å‹‰å¼·ã—ã¦å®Ÿè£…ã—ãŸã®ã§ã€ãã®å–ã‚Šçµ„ã¿ã®ç¶šãã¨ã—ã¦ã€ DCGAN(Deep Convolutional GAN(DCGAN)ã‚’å®Ÿè£…ã—ã¦éŠã‚“ã§ã¿ã‚‹ã€‚ ç”Ÿæˆçµæžœã¯ã“ã®ã‚ˆã†ã«ãªã£ãŸã€‚ (2017/9/7 è¿½è¨˜) DCGANã®è«–æ–‡ã‚’èªã‚“ã§ã¿ãŸã¨ã“ã‚ã€GANã®è«–æ–‡ã‚ˆã‚Šã‚‚èªã¿ã‚„ã™ã‹ã£ãŸã€‚ ã¾ãŸGANã®ã¨ãã«ã¯çœç•¥ã•ã‚Œã¦ã„ãŸãƒ¢ãƒ‡ãƒ«ã®æ§‹é€ ã‚‚æ›¸ã‹ã‚Œã¦ã„ãŸãŸã‚å®Ÿè£…ã®é›£æ˜“åº¦ãŒä½Žã‹ã£ãŸã€‚ ï¼ˆDCGANè‘—è€…ã®å®Ÿè£…ã‚‚å…¬é–‹ã•ã‚Œã¦ã„ã‚‹ãŸã‚ï¼Œãƒ‘ãƒ©ãƒ¡ãƒ¼ã‚¿ã‚’å‚è€ƒã«ã™ã‚‹ã“ã¨ã‚‚ã§ãã‚‹ï¼‰ DCGANã®è‚ã¯ä»¥ä¸‹ã®ä¸‰ç‚¹ã ï¼ˆã¨è«–æ–‡ã§è§¦ã‚Œã‚‰ã‚Œã¦ã„ã‚‹ï¼‰ deterministicãªpoolingæ‰‹æ³•ã®ä»£ã‚ã‚Šã«ã€ fractionally-stridedã‚’ä½¿ã£ã¦ã„ã‚‹ ã“ã¨ã€‚ã“ã‚Œã«ã‚ˆã‚Šã€ãƒãƒƒãƒˆãƒ¯ãƒ¼ã‚¯ãŒè‡ªèº«ã®downsamplingã‚’å¦ç¿’å¯èƒ½ã¨ãªã£ãŸã€‚ å…¨çµåˆå±¤ã‚’ä½¿ã£ã¦ã„ãªã„ ã“ã¨ã€‚convolutional featuresã‚’å…¥åŠ›å±¤ã¨å‡º
tsu-nera 2017/08/05
GAN
ãƒªãƒ³ã‚¯
Generative Adversarial Networks(GAN)ã‚’å‹‰å¼·ã—ã¦ã€kerasã§æ‰‹æ›¸ãæ–‡å—ç”Ÿæˆã™ã‚‹ - ç·‘èŒ¶æ€è€ƒãƒ–ãƒã‚°
Generative Adversarial Netsï¼ˆGANï¼‰ã¯ãƒ‹ãƒ¥ãƒ¼ãƒ©ãƒ«ãƒãƒƒãƒˆãƒ¯ãƒ¼ã‚¯ã®å¿œç”¨ã¨ã—ã¦ã€çµæ§‹ãªäººæ°—ãŒã‚ã‚‹ã€‚ãŸã¨ãˆã°Yann LeCunï¼ˆç¾åœ¨ã¯Facebookã«ã„ã‚‹ï¼‰ã¯GANã«ã¤ã„ã¦ä»¥ä¸‹ã®ã‚ˆã†ã«è¿°ã¹ã¦ã„ã‚‹ã€‚ â€œGenerative Adversarial Networks is the most interesting idea in the last ten years in machine learning.â€ GANã‚’å§‹ã‚ã¨ã™ã‚‹ç”Ÿæˆãƒ¢ãƒ‡ãƒ«ç³»ç ”ç©¶ã¯ã€ ã“ã‚Œã¾ã§äººé–“ã«ã—ã‹ã§ããªã„ã¨æ€ã‚ã‚Œã¦ã„ãŸã‚¯ãƒªã‚¨ã‚¤ãƒ†ã‚£ãƒ–ãªä»•äº‹ã«å¯¾ã—ã¦ã€ æ©Ÿæ¢°å¦ç¿’ãŒè¸ã¿è¾¼ã‚“ã§ã„ãæ§‹å›³ãªãŸã‚ã‹ã€å€‹äººçš„ã«ã‚‚ãƒ¯ã‚¯ãƒ¯ã‚¯ã™ã‚‹åˆ†é‡Žã ã€‚ ã“ã‚Œã¾ã§åˆ†é¡žå•é¡Œã‚’ä¸å¿ƒã«å®Ÿè£…ã—ã¦ãã¦ãã‚ãã‚é£½ãã¦ããŸãŸã‚ï¼Œ ä¸€ç•ªæœ€åˆã®GANè«–æ–‡ã‚’é ‘å¼µã£ã¦ç†è§£ã—ã¦ã€ ãã®å†…å®¹ã‚’kerasã§å®Ÿè£…ã—ã¦ã¿ã‚‹ã“ã¨ã«ã™ã‚‹ï¼Ž Generative Adver
tsu-nera 2017/08/05
GAN
ãƒªãƒ³ã‚¯
Wasserstein GANï¼ˆWGANï¼‰ã§ã„ã‚‰ã™ã¨ã‚„ç”»åƒã‚’ç”Ÿæˆã—ã¦ã¿ã‚‹ - ç·‘èŒ¶æ€è€ƒãƒ–ãƒã‚°
Wasserstein GANï¼Œç•¥ã—ã¦WGANã¨å‘¼ã°ã‚Œã‚‹ Arxiv: https://arxiv.org/abs/1611.02163 è‘—è€…ã«ã‚ˆã‚‹ã‚³ãƒ¼ãƒ‰: https://github.com/martinarjovsky/WassersteinGAN ã“ã®WGANã‚’ä½¿ã£ã¦ä»¥ä¸‹ã®ã‚ˆã†ãªã„ã‚‰ã™ã¨ã‚„ç”»åƒã‚’ç”Ÿæˆã—ãŸã¨ã„ã†ã®ãŒæœ¬è¨˜äº‹ã®ä¸»æ—¨ã€‚ æœ¬è«–æ–‡ã®contribution ç¢ºçŽ‡åˆ†å¸ƒé–“ã®è·é›¢ã‚’è¨ˆç®—ã™ã‚‹ãŸã‚ã®ï¼Œå¾“æ¥ä½¿ã‚ã‚Œã¦ããŸè·é›¢æŒ‡æ¨™ã¨Earth Mover(EM) distanceã¨ã®æ¯”è¼ƒã‚’è¡Œã£ãŸ EM distanceã®è¿‘ä¼¼é–¢æ•°ã‚’æœ€å°åŒ–ã™ã‚‹Wasserstein-GANã‚’å®šç¾©ã—ãŸ WGANãŒæ—¢å˜ã®GANã«ãŠã‘ã‚‹å•é¡Œã‚’è§£æ±ºã™ã‚‹ã“ã¨ã‚’ç¤ºã—ãŸ WGANã§ã¯discriminatorã¨generatorã®å¦ç¿’ãƒãƒ©ãƒ³ã‚¹ã‚’æ°—ã«ã—ãªãã¦è‰¯ã„ ãƒãƒƒãƒˆãƒ¯ãƒ¼ã‚¯æ§‹é€ ã‚‚æ°—ã«ã—ãªãã¦è‰¯ã„ mode dropping p
tsu-nera 2017/08/05
WGAN

GAN
ãƒªãƒ³ã‚¯
ã‚¿ã‚«ãƒã‚·æ˜¥ã® GAN ç¥ã‚Šï¼ã€œ ä¸€æ—¥ä¸€GAN(à¹‘â€¢Ì€ã…‚â€¢Ì)Ùˆâœ§ ã€œ - ABEJA Tech Blog
ABEJAã§ãƒªã‚µãƒ¼ãƒãƒ£ãƒ¼ã‚’ã—ã¦ã„ã¾ã™é«˜æ©‹ã§ã™ã€‚ æ˜¨ä»Š deep learning ç•Œéšˆã§ã¯ Generative Adversarial Net(GAN) ãŒæµè¡Œã£ã¦ã„ã¦ã€ä¸–ã¯ã¾ã•ã«ã‚¬ãƒ³ã‚¬ãƒ³è¡Œã“ã†ãœæ™‚ä»£ã§ã™ãã€‚ GAN ã‚’ç”¨ã„ã‚‹ã¨ç¶ºéº—ãªçµµãŒä½œæˆã§ããŸã‚ŠäºŒã¤ã®çµµã®ä¸é–“ã®ã‚ˆã†ãªçµµã‚’ç”Ÿæˆã§ããŸã‚Šã—ã¾ã™ã€‚ä¾‹ãˆã°ã“ã®è«–æ–‡ã®ã‚ˆã†ãªæ„Ÿã˜ã§ã™ã€‚ã“ã®ã‚ˆã†ã« GAN ã¯æœ‰ç”¨ãªãƒ¢ãƒ‡ãƒ«ã§ã‚ã‚‹ä¸€æ–¹ã€æœ€è¿‘ã® GAN ã§ã¯æ€¥ã«ã‚ˆãã‚ã‹ã‚‰ãªã„å¼ãŒå‡ºã¦ããŸã‚Šã™ã‚‹ã®ã§ã€å‹‰å¼·ã‚‚å…¼ãã¦ã€Œä¸€æ—¥ä¸€GANã€ã‚’ã‚„ã£ã¦ã¿ã¾ã—ãŸã€‚ä»Šå›žèªã‚“ã è«–æ–‡ã®ãƒªã‚¹ãƒˆã¯ä»¥ä¸‹ã§ã™ã€‚ EBGAN (https://arxiv.org/abs/1609.03126 ) WGAN (https://arxiv.org/abs/1701.07875) LSGAN (https://arxiv.org/abs/1611.04076) f-GAN (https://ar
tsu-nera 2017/08/05
GAN
ãƒªãƒ³ã‚¯
Wasserstein GAN (WGAN) - DeepLearningã‚’å‹‰å¼·ã™ã‚‹äºº
Wasserstein GAN (WGAN) [1701.07875] Wasserstein GAN ([1701.04862] Towards Principled Methods for Training Generative Adversarial Networks WGANã®è©±ã®å‰ã«ã“ã®è©±ãŒã‚ã‚‹) Martin Arjovskyæ°ã®å®Ÿè£…(Torch) GitHub - martinarjovsky/WassersteinGAN WGANã‚’TensorFlowã§å®Ÿè£…ã—ãŸ github.com è«–æ–‡ã¾ã¨ã‚ æ·±ãç†è§£ã—ã‚ˆã†ã¨ã™ã‚‹ã¨ã‹ãªã‚Šæ•°å¦çš„çŸ¥è˜ãŒè¦æ±‚ã•ã‚Œã‚‹ã®ã§ã€ç›´æ„Ÿçš„ãªç†è§£ã ã‘ã‚’è¿½ã£ã¦ã„ãã“ã¨ã«ã™ã‚‹. ãã‚‚ãã‚‚VAEã‚„GANã®ã‚¢ãƒ—ãƒãƒ¼ãƒã¨ã„ã†ã®ã¯ã€è‡ªåˆ†ã®åˆ†å¸ƒã‚’çœŸã®åˆ†å¸ƒã«è¿‘ã¥ã‘ã¦ã„ãã¨ã„ã†ã‚‚ã®ã ãŒã€ã“ã®åˆ†å¸ƒé–“ã®è·é›¢/ãƒ€ã‚¤ãƒãƒ¼ã‚¸ã‚§ãƒ³ã‚¹ã®å®šç¾©ã®ä»•æ–¹ã«ã‚ˆã£ã¦ã€å¦ç¿’ã®åŽæŸæ€§ã€å®‰å®šæ€§ã«å·®ç•°ãŒã‚ã‚‹
tsu-nera 2017/08/05
WGAN

GAN
ãƒªãƒ³ã‚¯
Wasserstein GAN [arXiv:1701.07875]
æ¦‚è¦ Wasserstein GANã‚’èªã‚“ã Chainerã§å®Ÿè£…ã—ãŸ ã¯ã˜ã‚ã« Wasserstein GANï¼ˆä»¥ä¸‹WGANï¼‰ã¯Earth Moverâ€™s Distanceï¼ˆã¾ãŸã¯Wasserstein Distanceï¼‰ã‚’æœ€å°åŒ–ã™ã‚‹å…¨ãæ–°ã—ã„GANã®å¦ç¿’æ–¹æ³•ã‚’ææ¡ˆã—ã¦ã„ã¾ã™ã€‚ å®Ÿè£…ã«ã‚ãŸã£ã¦äº‹å‰çŸ¥è˜ã¯ä¸è¦ã§ã™ã€‚ ç§ã¯Earth Moverâ€™s Distanceï¼ˆEDMï¼‰ãªã©ã‚’äº‹å‰ã«èª¿ã¹ã¦ãŠãã¾ã—ãŸãŒå®Ÿè£…ã«é–¢ä¿‚ã‚ã‚Šã¾ã›ã‚“ã§ã—ãŸã€‚ ã¾ãŸRedditã®WGANã®ã‚¹ãƒ¬ãƒƒãƒ‰ã«ã¦ã€GANã®è€ƒæ¡ˆè€…ã§ã‚ã‚‹Ian Goodfellowæ°ã‚„æœ¬è«–æ–‡ã®è‘—è€…Martin Arjovskyæ°ãŒæ´»ç™ºã«è°è«–ã‚’äº¤ã‚ã—ã¦ã„ã¾ã™ã€‚ Martin Arjovskyæ°ã®å®Ÿè£…ãŒGithubã§å…¬é–‹ã•ã‚Œã¦ã„ã¾ã™ã®ã§å®Ÿè£…ã«ã¯å›°ã‚‰ãªã„ã¨æ€ã„ã¾ã™ã€‚ ç§ã¯Chainer 1.20ã§å®Ÿè£…ã—ã¾ã—ãŸã€‚ https://github.com/mus
tsu-nera 2017/08/05
WGAN

GAN
ãƒªãƒ³ã‚¯
GitHub - soumith/ganhacks: starter from "How to Train a GAN?" at NIPS2016
You signed in with another tab or window. Reload to refresh your session. You signed out in another tab or window. Reload to refresh your session. You switched accounts on another tab or window. Reload to refresh your session. Dismiss alert
tsu-nera 2017/08/05
GAN Hacks

GAN
ãƒªãƒ³ã‚¯
1