SlideShare a Scribd company logo

[213]building ai to recreate our visual world

•
NAVER D2, profile picture
NAVER D2

The document discusses recent advances in generative models for visual data synthesis and translation. It provides an overview of generative adversarial networks (GANs) and their applications, including generating images from random noise and translating images from one domain to another. Cycle-consistent adversarial networks (CycleGAN) are introduced as a way to perform image-to-image translation without paired training examples. The document highlights several examples of visual synthesis and translation using GANs and CycleGAN, including style transfer, domain adaptation, and generating images from segmentation maps.

Technology
1 of 135
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
Learning to recreate our visual world Jun-Yan Zhu UC Berkeley
4.7 trillion photographs 13 billion images 300 million images uploaded daily 1.5 million images uploaded daily 300 hours uploaded per minute
Street Visual Understanding Scene understanding [Zhou et al. 2014]
people bicycle umbrella Visual Understanding Object Detection [Girshick et al. 2014]
A city street filled with lots of people walking in the rain. Visual Understanding [Karpathy and Fei-Fei 2015] [Donahue et al. 2015] [Misra et al. 2015] Image Captioning
A city street filled with lots of people walking in the rain. Visual Understanding Visual Synthesis
A city street filled with lots of people walking in the rain. Visual Understanding Visual Synthesis by Dom’s father
Man-Computer Symbiosis Computers can help us draw, even if we can’t. JCR Licklider, 1968
Sketchpad Sutherland, 1960
Image warping By Photoshop Visual Synthesis is hard! Why it does not work - Is this a handbag? - What makes a handbag look real?
What makes an image look real? Statistics of Natural Images and Models [Huang and Mumford. 2000] Sample images from the dataset (4000 images in total) log histogram of pixels log histogram of wavelet pairs
What makes an image look fake? Mismatch color statistics [Lalonde and Efros 2007] lack details [Johnson et al. 2011] noticeable boundary [Perez et al. 2003] “Bleeding” artifacts [Tao et al. 2010]
Deep Generative Models • Generative Adversarial Network (GAN) [Goodfellow et al. 2014] • Variational Auto-Encoder (VAE) [Kingma and Welling 2013] • DRAW (Recurrent Neural Network) [Gregor et al 2015] • Pixel RNN and Pixel CNN ([Oord et al 2016]) • …
Generative Adversarial Networks (GANs) Discriminato r Real vs. Fake Generator 𝑥 ~ 𝐺(𝑧) [Goodfellow et al. 2014]
Generator [Goodfellow et al. 2014] z G(z) cat credit: aleju/cat-generator G
real or fake? [Goodfellow et al. 2014] Discriminator z G(z) D Generator G 𝐺: generate fake samples that can fool 𝐷 𝐷: classify fake samples vs. real images
[Goodfellow et al. 2014]
[Goodfellow et al. 2014] fake 0.1 z G(z) DG
real 0.9 [Goodfellow et al. 2014] z G(z) DG D x fake 0.1
fake 0.3 [Goodfellow et al. 2014] z G(z) DG D x real 0.9
Images generated by GANs Sample shoes images from Zappos.com [Yu and Grauman 2014] Random image samples from Generator G(z) [Radford et al. 2015]
Explore the GANs latent space [Radford et al. 2015] 𝐺(𝑧-) 𝐺(𝑧.)Linear Interpolation in z space: 𝐺(𝑧- + 𝑡 ⋅ (𝑧. − 𝑧-)) Handbag model trained on 137k Amazon handbags
Limitations of GANs • produce images randomly; hard to control. • not photo-realistic enough; low resolution.
Limitations of GANs • produce images randomly; hard to control. • not photo-realistic enough; low resolution.
Controllable Image Generation [Zhu et al. ECCV 2016] Trained on Places Dataset
Generative Visual Manipulation on the Natural Image Manifold Zhu, Krähenbühl, Shechtman, Efros 2016
Manipulating the Latent Vector Objective: user guidance imageconstraint violation loss 𝐿4 Generative model: 𝐺(𝑧) z G(z) G 𝐿4( ) , 𝑣4
Manipulating the Latent Vector Objective: user guidance imageconstraint violation loss 𝐿4 Generative model: 𝐺(𝑧) z G(z) G 𝐿4( ) , 𝑣4
Manipulating the Latent Vector z G(z) G 𝐿4( ) , 𝑣4 𝐺(𝑧) Guidance 𝑣4
Product Design
Overview original photo projection on manifold Project Results generated by GANs Editing UI
Overview original photo projection on manifold Project Edit Transfer Results generated by GANs Final results Editing UI
Edit Transfer via Generalized Optical Flow 𝐺(𝑧.)𝐺(𝑧-) Input Motion (u, v)+ Color (𝑨 𝟑×𝟒): estimate per-pixel geometric and color variation Linear Interpolation in 𝑧 space [Brox et al. 2004] [Shih et al. 2013]
Edit Transfer via Generalized Optical Flow 𝐺(𝑧.)𝐺(𝑧-) Input Linear Interpolation in 𝑧 space Motion (u, v)+ Color (𝑨 𝟑×𝟒): estimate per-pixel geometric and color variation Motion (u, v)+ Color (𝑨 𝟑×𝟒): estimate per-pixel geometric and color variation
Edit Transfer via Generalized Optical Flow Result 𝐺(𝑧.)𝐺(𝑧-) Input Motion (u, v)+ Color (𝑨 𝟑×𝟒): estimate per-pixel geometric and color variation Linear Interpolation in 𝑧 space
Editing handbags
Editing shoes
Generated Image User Guidance Scribble→ latent space → Image [Zhu et al. 2016] G Scribble ImageLatent space
Generated Image User Guidance Scribble→ latent space → Image [Zhu et al. 2016] Image-to-Image Network [Isola, Zhu, Zhou, Efros. 2017] [Zhu*, Park*, Isola, Efros. 2017]
Image colorization Designing objective functions L2 regression [Johnson et al. 2016] Super-resolution [Zhang et al. 2016] L2 regression slides credit: Phillip Isola
Image colorization Designing objective functions Cross entropy objective, with colorfulness term Deep feature covariance matching objective [Johnson et al. 2016] Super-resolution [Zhang et al. 2016]
Universal loss? … …
… Generated vs Real (classifier) Real photos Generated images … … [Goodfellow et al. 2014]
Image-to-Image Translation Image-to-image translation with conditional adversarial nets [Isola, Zhu, Zhou, Efros. CVPR 2017]
real or fake? [Goodfellow et al. 2014] Discriminator z G(z) D Generator G 𝐺: generate fake samples that can fool 𝐷 𝐷: classify fake samples vs. real images
real or fake? [Goodfellow et al. 2014] Discriminator x G(x) D Generator G 𝐺: generate fake samples that can fool 𝐷 𝐷: classify fake samples vs. real images
Real! [Goodfellow et al. 2014] Discriminator x G(x) D Generator G
[Goodfellow et al. 2014] Discriminator x G(x) D Generator G Real too!
real or fake pair ? slides credit: Many slides modified from Phillip Isola’s talk x G(x) G D
fake pair: 0.1 G D x G(x)
real pair: 0.9 𝑦 D
Edges → Images Input Output Input Output Input Output Edges from [Xie & Tu, 2015]
Sketches → Images Input Output Input Output Input Output Trained on Edges → Images Data from [Eitz, Hays, Alexa, 2012]
#edges2cats [Christopher Hesse] Ivy Tasi @ivymyt @gods_tail @matthematician https://affinelayer.com/pixsrv/ Vitaly Vidmirov @vvid
Input Output Groundtruth Data from [maps.google.com]
Input Output Groundtruth Data from [maps.google.com]
Labels → Facades Input Output Input Output Data from [Tylecek, 2013]
BW → Color Input Output Input Output Input Output Data from [Russakovsky et al. 2015]
… Paired
- Expensive to collect pairs. - Impossible in many scenarios. Label ↔ photo: per-pixel labeling … Paired Horse ↔ zebra: how to get zebras?
… … … Paired Unpaired
x G(x) Generator G D No input-output pairs!
Discriminator x G(x) D Generator G Real!
Discriminator x G(x) D Generator G Real too! GANs doesn’t force output to correspond to input
mode collapse!
…CycleGAN, or there and back aGAN [Zhu*, Park*, Isola, Efros. ICCV 2017] …… Discriminator D>: 𝐿?@A 𝐺 𝑥 , 𝑦 Real zebras vs. fake zebras
Discriminator DB: 𝐿?@A 𝐹 𝑦 , 𝑥 Real horses vs. fake horses Discriminator D>: 𝐿?@A 𝐺 𝑥 , 𝑦 Real zebras vs. fake zebras … …CycleGAN, or there and back aGAN
Cycle-consistency Loss Forward cycle loss: F G x − x . G(x) F(G x )x
Cycle-consistency Loss Large cycle loss Forward cycle loss: F G x − x . G(x) F(G x )x Small cycle loss
Cycle-consistency Loss Backward cycle loss: 𝐺 𝐹 𝑦 − 𝑦 . Forward cycle loss: F G x − x . G(x) F(G x )x F(y) G(F x )𝑦 See also [Yi et al. 2017] [Kim et al. 2017]
Results
Collection Style Transfer Photograph @ Alexei Efros MonetVan Gogh CezanneUkiyo-e
Cezanne Ukiyo-eMonetInput Van Gogh
Monet’s paintings → photos
Monet’s paintings → photos
[213]building ai to recreate our visual world
[213]building ai to recreate our visual world
[213]building ai to recreate our visual world
[213]building ai to recreate our visual world
Neural Style Transfer [Gatys et al. 2015]
Input Style Image I CycleGANStyle image II Entire collection Photo → Van Gogh horse → zebra Input Style image I CycleGANStyle image II Entire collection
CG2Real: GTA5 → real streetview Inspired by [Johnson et al. 2011]GTA5 CG Input Output
Real2CG: real streetview → GTA Cityscape Input Output
[213]building ai to recreate our visual world
[Richter*, Vineet* et al. 2016] GTA5 images Segmentation labels
Use CG data to train recognition systems Per-class accuracy Per-pixel accuracy Oracle (Train and test on Real) 60.3 93.1 Train on ‘free’ synthetic data (GTA5) Test on real images meanIOU (per-class) Per-pixel accuracy Oracle (Train and test on Real) 60.3 93.1 Train on CG, test on Real 17.9 54.0
Domain Adaption with CycleGAN [Tzeng et al. ] In submission meanIOU (per-class) Per-pixel accuracy Oracle (Train and test on Real) 60.3 93.1 Train on CG, test on Real 17.9 54.0 FCN in the wild [Hoffman et al.] 27.1 (+6.0) - Train on CycleGAN, test on Real 34.8 (+16.9) 82.8 Train on CycleGAN data Test on real images
Failure cases
[213]building ai to recreate our visual world
[213]building ai to recreate our visual world
ImageNet “Wild horse”
Conclusion
A city street filled with lots of people walking in the rain. Visual Understanding Visual Synthesis by Dom’s father
A city street filled with lots of people walking in the rain. Visual Understanding Visual Synthesis by iGAN
A city street filled with lots of people walking in the rain. Visual Understanding Visual Synthesis by Pix2pix
A city street filled with lots of people walking in the rain. Visual Understanding Visual Synthesis by CycleGAN
Visual Synthesis A few things I have learned • Visual Synthesis is a learning problem. • We can learn to do it with trillions of photos. • Build general-purpose tools and find cool problems. • Open-source the code and data.
Phillip Isola and Jun-Yan Zhu
Community-driven research: #pix2pix Bertrand Gondouin @bgondouinBrannon Dorsey @brannondorsey Kaihu Chen @kaihuchen Mario Klingemann @quasimondo
Jun-Yan Zhu and Taesung Park
Pix2pix: 144 lines CycleGAN: 220 lines Jun-Yan Zhu and Taesung Park
20+ Implementations by others • [Tensorflow] (by Harry Yang) • [Tensorflow] (by Archit Rathore) • [Tensorflow] (by Van Huy) • [Tensorflow] (by Xiaowei Hu) • [Tensorflow-simple] (by Zhenliang He) • [Chainer] (by Yanghua Jin) • [Minimal PyTorch] (by yunjey) • [Mxnet] (by Ldpe2G) • [lasagne/keras] (by tjwei) ⋯
Birds @Matt Powell Bear → Panda @Matt Powell #CycleGAN Monet → Thomas Kinkade @David Fouhey Resurrecting Ancient Cities @ Jack Clark
Portrait to Dollface @Mario Klingemann Colorizing legacy photographs @Mario Klingemann #CycleGAN
Face Swapping with #CycleGAN Input
Face Swapping with #CycleGAN Input Generated
Face Swapping with #CycleGAN Input Generated Reconstruction
#CycleGAN: Face ↔ Ramen @ Takuya Kato
@itok_msi + Smaller cycle loss + Global image discriminator
Medical imaging appilcations • Segmentation [Xue et al.], etc. • CT denoising [Yi and Babyn] • MR <-> CT [Wolterink et al. ] • MRI Reconstruction [Quan et al. ] ⋯
MR <-> CT
MRI Reconstruction
Three views of GANs • generative models G 𝑧 unsupervised learning • Trainable regression loss. • Domain matching loss. “What should I do”
Thank You! Eli ShechtmanPhilipp Krähenbühl Alyosha Efros Yong Jae LeePhillip Isola Taesung Park Richard Zhang Tinghui ZhouTing-Chun Wang Deepak Pathak Trevor DarrellRavi Ramamoorthi Nima K. Kalantari Jue WangOliver Wang Aseem Agarwala Eric Tzeng Ce Liu Miki Rubinstein Judy Hoffman Manmohan Chandraker Young GengAngela Lin Aquarius
Questions? @LynnHo
Domain adaption: train on source, adapt to target • A recent survey: ~300 papers • Minimizing distribution distance – Borgwardt 06, Mansour 09, Pan 09, Fernando 13 • Deep model adaption – Chopra 13, Tzeng 14, Long 15, Ganin 15, Hoffman 17. Slides credit: Judy Hoffman
State-of-the-art domain adaption method Per-class accuracy Per-pixel accuracy Oracle (Train and test on Real) 60.3 93.1 Train on CG, test on Real 17.9 54.0 FCN in the wild [Hoffman et al.] 27.1 (+6.0) - VGG-FCN: 17.9 [Long et al. 2015]; Dilated-VGG-FCN: 21.1 [Fisher and Koltun 2015’] Train on ‘free’ synthetic data (GTA5) Test on real images adapt
Cats are as popular as GANs • GitHub: github.com/junyanz/CatPapers • 90% data is visual; most of visual data are about Cats. • 70+ vision, learning and graphics papers.
[213]building ai to recreate our visual world
Most influential cat paper • Fred Attneave. “Some informational aspects of visual perception”. Psychological Review (1954).
Questions?
User-Guided Colorization [Zhang*, Zhu* et al. SIGGRAPH 17] Grayscale image Output colorization User colors, mask Raw Data
Training Details: Objective function • Conditional GAN
Training Details: Objective function • Conditional GAN + L1 𝑦 , { } • Stable training + fast convergence.
Training Details: Discriminator 𝐷 [Radford et al., 2015] ImageGAN Slide from Victor Garcia • Faster, fewer parameters; Arbitrarily large images • Equal or better results Our discriminator PatchGAN
Effects of discriminator 𝐷
Training Details: Generator 𝐺 Encoder-decoder U-Net [Ronneberger et al.]
Shallower depth of field
iPhone DSLR iPhone DSLR
Summer ↔ Winter
Ablation study on paired dataset the same output mode collapse!
Ablation study on paired dataset the same output mode collapse!
Ablation Study on Cityscapes dataset
Training Details: Objective • GANs with cross-entropy loss • Least square GANs [Mao et al. 2016] Stable training + better results Vanishing gradients
Training Details: Generator 𝐺 Encoder-decoderU-Net [Ronneberger et al.] ResNet [He et al.] [Johnson et al.] • Both have skip connections • ResNet: fewer parameters better for ill-posed problems

More Related Content

PDF
[231]운영체제 수준에서의 데이터베이스 성능 분석과 최적화
NAVER D2
 
PDF
[234]멀티테넌트 하둡 클러스터 운영 경험기
NAVER D2
 
PDF
[241]large scale search with polysemous codes
NAVER D2
 
PDF
[215]streetwise machine learning for painless parking
NAVER D2
 
PDF
[223]rye, 샤딩을 지원하는 오픈소스 관계형 dbms
NAVER D2
 
PDF
[242]open stack neutron dataplane 구현
NAVER D2
 
PDF
[222]neural machine translation (nmt) 동작의 시각화 및 분석 방법
NAVER D2
 
PDF
[224]nsml 상상하는 모든 것이 이루어지는 클라우드 머신러닝 플랫폼
NAVER D2
 
[231]운영체제 수준에서의 데이터베이스 성능 분석과 최적화
NAVER D2
 
[234]멀티테넌트 하둡 클러스터 운영 경험기
NAVER D2
 
[241]large scale search with polysemous codes
NAVER D2
 
[215]streetwise machine learning for painless parking
NAVER D2
 
[223]rye, 샤딩을 지원하는 오픈소스 관계형 dbms
NAVER D2
 
[242]open stack neutron dataplane 구현
NAVER D2
 
[222]neural machine translation (nmt) 동작의 시각화 및 분석 방법
NAVER D2
 
[224]nsml 상상하는 모든 것이 이루어지는 클라우드 머신러닝 플랫폼
NAVER D2
 

Viewers also liked (20)

PDF
[221]똑똑한 인공지능 dj 비서 clova music
NAVER D2
 
PDF
[212]big models without big data using domain specific deep networks in data-...
NAVER D2
 
PDF
[246]reasoning, attention and memory toward differentiable reasoning machines
NAVER D2
 
PDF
[225]빅데이터를 위한 분산 딥러닝 플랫폼 만들기
NAVER D2
 
PDF
[232]mist 고성능 iot 스트림 처리 시스템
NAVER D2
 
PDF
인공지능추천시스템 airs개발기_모델링과시스템
NAVER D2
 
PPTX
[213] 의료 ai를 위해 세상에 없는 양질의 data 만드는 도구 제작하기
NAVER D2
 
PDF
백억개의 로그를 모아 검색하고 분석하고 학습도 시켜보자 : 로기스
NAVER D2
 
PPTX
[211] HBase 기반 검색 데이터 저장소 (공개용)
NAVER D2
 
PDF
유연하고 확장성 있는 빅데이터 처리
NAVER D2
 
PDF
[216]네이버 검색 사용자를 만족시켜라! 의도파악과 의미검색
NAVER D2
 
PPTX
[244]네트워크 모니터링 시스템(nms)을 지탱하는 기술
NAVER D2
 
PDF
[141]네이버랩스의 로보틱스 연구 소개
NAVER D2
 
PDF
웨일브라우저 성능 및 메모리 최적화
NAVER D2
 
PDF
밑바닥부터시작하는360뷰어
NAVER D2
 
PDF
[131]chromium binging 기술을 node.js에 적용해보자
NAVER D2
 
PDF
[124]자율주행과 기계학습
NAVER D2
 
PDF
[111]open, share, enjoy 네이버의 오픈소스 활동
NAVER D2
 
PDF
[112]clova platform 인공지능을 엮는 기술
NAVER D2
 
PDF
[142] 생체 이해에 기반한 로봇 – 고성능 로봇에게 인간의 유연함과 안전성 부여하기
NAVER D2
 
[221]똑똑한 인공지능 dj 비서 clova music
NAVER D2
 
[212]big models without big data using domain specific deep networks in data-...
NAVER D2
 
[246]reasoning, attention and memory toward differentiable reasoning machines
NAVER D2
 
[225]빅데이터를 위한 분산 딥러닝 플랫폼 만들기
NAVER D2
 
[232]mist 고성능 iot 스트림 처리 시스템
NAVER D2
 
인공지능추천시스템 airs개발기_모델링과시스템
NAVER D2
 
[213] 의료 ai를 위해 세상에 없는 양질의 data 만드는 도구 제작하기
NAVER D2
 
백억개의 로그를 모아 검색하고 분석하고 학습도 시켜보자 : 로기스
NAVER D2
 
[211] HBase 기반 검색 데이터 저장소 (공개용)
NAVER D2
 
유연하고 확장성 있는 빅데이터 처리
NAVER D2
 
[216]네이버 검색 사용자를 만족시켜라! 의도파악과 의미검색
NAVER D2
 
[244]네트워크 모니터링 시스템(nms)을 지탱하는 기술
NAVER D2
 
[141]네이버랩스의 로보틱스 연구 소개
NAVER D2
 
웨일브라우저 성능 및 메모리 최적화
NAVER D2
 
밑바닥부터시작하는360뷰어
NAVER D2
 
[131]chromium binging 기술을 node.js에 적용해보자
NAVER D2
 
[124]자율주행과 기계학습
NAVER D2
 
[111]open, share, enjoy 네이버의 오픈소스 활동
NAVER D2
 
[112]clova platform 인공지능을 엮는 기술
NAVER D2
 
[142] 생체 이해에 기반한 로봇 – 고성능 로봇에게 인간의 유연함과 안전성 부여하기
NAVER D2
 
Ad

More from NAVER D2 (20)

PDF
[211] 인공지능이 인공지능 챗봇을 만든다
NAVER D2
 
PDF
[233] 대형 컨테이너 클러스터에서의 고가용성 Network Load Balancing: Maglev Hashing Scheduler i...
NAVER D2
 
PDF
[215] Druid로 쉽고 빠르게 데이터 분석하기
NAVER D2
 
PDF
[245]Papago Internals: 모델분석과 응용기술 개발
NAVER D2
 
PDF
[236] 스트림 저장소 최적화 이야기: 아파치 드루이드로부터 얻은 교훈
NAVER D2
 
PDF
[235]Wikipedia-scale Q&A
NAVER D2
 
PDF
[244]로봇이 현실 세계에 대해 학습하도록 만들기
NAVER D2
 
PDF
[243] Deep Learning to help student’s Deep Learning
NAVER D2
 
PDF
[234]Fast & Accurate Data Annotation Pipeline for AI applications
NAVER D2
 
PDF
Old version: [233]대형 컨테이너 클러스터에서의 고가용성 Network Load Balancing
NAVER D2
 
PDF
[226]NAVER 광고 deep click prediction: 모델링부터 서빙까지
NAVER D2
 
PDF
[225]NSML: 머신러닝 플랫폼 서비스하기 & 모델 튜닝 자동화하기
NAVER D2
 
PDF
[224]네이버 검색과 개인화
NAVER D2
 
PDF
[216]Search Reliability Engineering (부제: 지진에도 흔들리지 않는 네이버 검색시스템)
NAVER D2
 
PDF
[214] Ai Serving Platform: 하루 수 억 건의 인퍼런스를 처리하기 위한 고군분투기
NAVER D2
 
PDF
[213] Fashion Visual Search
NAVER D2
 
PDF
[232] TensorRT를 활용한 딥러닝 Inference 최적화
NAVER D2
 
PDF
[242]컴퓨터 비전을 이용한 실내 지도 자동 업데이트 방법: 딥러닝을 통한 POI 변화 탐지
NAVER D2
 
PDF
[212]C3, 데이터 처리에서 서빙까지 가능한 하둡 클러스터
NAVER D2
 
PDF
[223]기계독해 QA: 검색인가, NLP인가?
NAVER D2
 
[211] 인공지능이 인공지능 챗봇을 만든다
NAVER D2
 
[233] 대형 컨테이너 클러스터에서의 고가용성 Network Load Balancing: Maglev Hashing Scheduler i...
NAVER D2
 
[215] Druid로 쉽고 빠르게 데이터 분석하기
NAVER D2
 
[245]Papago Internals: 모델분석과 응용기술 개발
NAVER D2
 
[236] 스트림 저장소 최적화 이야기: 아파치 드루이드로부터 얻은 교훈
NAVER D2
 
[235]Wikipedia-scale Q&A
NAVER D2
 
[244]로봇이 현실 세계에 대해 학습하도록 만들기
NAVER D2
 
[243] Deep Learning to help student’s Deep Learning
NAVER D2
 
[234]Fast & Accurate Data Annotation Pipeline for AI applications
NAVER D2
 
Old version: [233]대형 컨테이너 클러스터에서의 고가용성 Network Load Balancing
NAVER D2
 
[226]NAVER 광고 deep click prediction: 모델링부터 서빙까지
NAVER D2
 
[225]NSML: 머신러닝 플랫폼 서비스하기 & 모델 튜닝 자동화하기
NAVER D2
 
[224]네이버 검색과 개인화
NAVER D2
 
[216]Search Reliability Engineering (부제: 지진에도 흔들리지 않는 네이버 검색시스템)
NAVER D2
 
[214] Ai Serving Platform: 하루 수 억 건의 인퍼런스를 처리하기 위한 고군분투기
NAVER D2
 
[213] Fashion Visual Search
NAVER D2
 
[232] TensorRT를 활용한 딥러닝 Inference 최적화
NAVER D2
 
[242]컴퓨터 비전을 이용한 실내 지도 자동 업데이트 방법: 딥러닝을 통한 POI 변화 탐지
NAVER D2
 
[212]C3, 데이터 처리에서 서빙까지 가능한 하둡 클러스터
NAVER D2
 
[223]기계독해 QA: 검색인가, NLP인가?
NAVER D2
 
Ad

Recently uploaded (20)

PDF
Orbitly Pitch Deck｜A Mission-Driven Platform for Side Project Collaboration (...
zz41354899
 
PDF
Oracle AI Vector Search- Getting Started and what's new in 2025- AIOUG Yatra ...
Sandesh Rao
 
PPTX
IoT Sensor Integration 2025 Powering Smart Tech and Industrial Automation.pptx
Rejig Digital
 
PPTX
The-Ethical-Hackers-Imperative-Safeguarding-the-Digital-Frontier.pptx
sujalchauhan1305
 
PDF
Brief History of Internet - Early Days of Internet
sutharharshit158
 
PDF
Software Development Company | KodekX
KodekX
 
PDF
Tea4chat - another LLM Project by Kerem Atam
a0m0rajab1
 
PPTX
OA presentation.pptx OA presentation.pptx
pateldhruv002338
 
PDF
The Evolution of KM Roles (Presented at Knowledge Summit Dublin 2025)
Enterprise Knowledge
 
PPTX
Coupa-Overview _Assumptions presentation
annapureddyn
 
PDF
AI Unleashed - Shaping the Future -Starting Today - AIOUG Yatra 2025 - For Co...
Sandesh Rao
 
PDF
Structs to JSON: How Go Powers REST APIs
Emily Achieng
 
PPTX
AI and Robotics for Human Well-being.pptx
JAYMIN SUTHAR
 
PDF
Automating ArcGIS Content Discovery with FME: A Real World Use Case
Safe Software
 
PDF
Data_Analytics_vs_Data_Science_vs_BI_by_CA_Suvidha_Chaplot.pdf
CA Suvidha Chaplot
 
PPTX
Applied-Statistics-Mastering-Data-Driven-Decisions.pptx
parmaryashparmaryash
 
PDF
Security features in Dell, HP, and Lenovo PC systems: A research-based compar...
Principled Technologies
 
PDF
This slide provides an overview Technology
mineshkharadi333
 
PDF
How Open Source Changed My Career by abdelrahman ismail
a0m0rajab1
 
PDF
Beyond Automation: The Role of IoT Sensor Integration in Next-Gen Industries
Rejig Digital
 
Orbitly Pitch Deck｜A Mission-Driven Platform for Side Project Collaboration (...
zz41354899
 
Oracle AI Vector Search- Getting Started and what's new in 2025- AIOUG Yatra ...
Sandesh Rao
 
IoT Sensor Integration 2025 Powering Smart Tech and Industrial Automation.pptx
Rejig Digital
 
The-Ethical-Hackers-Imperative-Safeguarding-the-Digital-Frontier.pptx
sujalchauhan1305
 
Brief History of Internet - Early Days of Internet
sutharharshit158
 
Software Development Company | KodekX
KodekX
 
Tea4chat - another LLM Project by Kerem Atam
a0m0rajab1
 
OA presentation.pptx OA presentation.pptx
pateldhruv002338
 
The Evolution of KM Roles (Presented at Knowledge Summit Dublin 2025)
Enterprise Knowledge
 
Coupa-Overview _Assumptions presentation
annapureddyn
 
AI Unleashed - Shaping the Future -Starting Today - AIOUG Yatra 2025 - For Co...
Sandesh Rao
 
Structs to JSON: How Go Powers REST APIs
Emily Achieng
 
AI and Robotics for Human Well-being.pptx
JAYMIN SUTHAR
 
Automating ArcGIS Content Discovery with FME: A Real World Use Case
Safe Software
 
Data_Analytics_vs_Data_Science_vs_BI_by_CA_Suvidha_Chaplot.pdf
CA Suvidha Chaplot
 
Applied-Statistics-Mastering-Data-Driven-Decisions.pptx
parmaryashparmaryash
 
Security features in Dell, HP, and Lenovo PC systems: A research-based compar...
Principled Technologies
 
This slide provides an overview Technology
mineshkharadi333
 
How Open Source Changed My Career by abdelrahman ismail
a0m0rajab1
 
Beyond Automation: The Role of IoT Sensor Integration in Next-Gen Industries
Rejig Digital
 

[213]building ai to recreate our visual world