Emer Emergenc ence o e of O f Obj bjec ect S Seg egmen - PowerPoint PPT Presentation

Emer Emergenc ence o e of O f Obj bjec ect S Seg egmen mentatio ion n in P in Per ertur urbed Gener bed Generativ ive M e Models dels Adam Bielski, Paolo Favaro University of Bern

Ob Object se segm gmentation Segmentation model 2

Ob Object se segm gmentation Annotated data Segmentation model 3

Ob Object se segm gmentation Annotated data Segmentation model • Human annotated • Expensive • Expertise knowledge 4

Un Unsuper ervis vised ed objec ject t seg egmen entatio tion Annotated data Segmentation model • Human annotated • Expensive • Expertise knowledge Can we do it without labels? 5

Un Unsuper ervis vised ed ob object se segm gmentation E G Mask Image 6

Le Learn rning to to ge generate la layer ered ed sc scenes wi with GA GAN Background Foreground E G Mask Image 𝑨~𝒪(0, 𝐽) 7

Le Learn rning to to ge generate la layer ered ed sc scenes wi with GA GAN Background Image composition Foreground + ⊙ ⊙ = E G Background 1 - Mask Foreground Mask Mask Composite image Image 𝑨~𝒪(0, 𝐽) 8

Le Learn rning to to ge generate la layer ered ed sc scenes wi with GA GAN Background Image composition Foreground + ⊙ ⊙ = E G Background 1 - Mask Foreground Mask Mask Composite image Image 𝑨~𝒪(0, 𝐽) Real / D Fake Unlabeled real images 9

De Degenerate s solu lutio ions Background Foreground Mask Composite image 10

Wha What is is a c a correct ect par partit itio ion? n? 14

Wha What is is a c a correct ect par partit itio ion? n? 15

Ra Rando ndom m shi hifts 16

Invalid partition ⟹ invalid scene after a small shift Valid partition ⟹ valid scene after a small shift Ra Rando ndom m shi hifts 20

Av Avoiding degenerate solutions Background Foreground Mask Composite image Composite image w/shift 21

Av Avoiding degenerate solutions Background Foreground Mask Composite image Composite image w/shift Add a loss term to avoid empty masks 25

Av Avoiding degenerate solutions Background Foreground Mask Composite image Composite image w/shift Add a loss term to avoid empty masks 26

Learn Le rning wi with per pertur urba bations ns Background Image composition Foreground + ⊙ ⊙ = E G Background 1 - Mask Foreground Mask Mask Composite image Image 𝑨~𝒪(0, 𝐽) Minimum mask loss 27

Learn Le rning wi with per pertur urba bations ns Background Image composition Foreground + ⊙ ⊙ = E G Background 1 – Mask Foreground Mask Mask Composite image Image shifted shifted shifted (fake) 𝑨~𝒪(0, 𝐽) Random shift 𝑞~𝒱 Real Minimum mask loss / D Fake Real image 28

Ge Generatio tion results lts LSUN Car LSUN Bird LSUN Chair LSUN Horse Backg�o�nd Backg�o�nd Fo�eg�o�nd Fo�eg�o�nd Ma�k Ma�k Fo�eg�o�nd Fo�eg�o�nd �/�ma�k �/�ma�k Com�o�i�e Com�o�i�e image image 29

Le Learn rning to to se segm gment Background Image composition Foreground + ⊙ ⊙ = G Background 1 - Mask Mask Mask Foreground Composite image 30

Learn Le rning to to se segm gment Frozen Background Image composition Foreground + ⊙ ⊙ = G Background 1 - Mask Mask Mask Foreground Composite image 31

Le Learn rning to to se segm gment Frozen Background Image composition Foreground + ⊙ ⊙ = E G Background 1 - Mask Mask Mask Foreground Composite image Image 𝑀1 𝑀𝑝𝑡𝑡 32

Le Learn rning to to se segm gment Frozen Background Image composition Foreground + ⊙ ⊙ = E G Background 1 - Mask Mask Mask Foreground Composite image Image 𝑀1 𝑀𝑝𝑡𝑡 33

Se Segme mentation on r results Real image Real image O�� Our results me�hod Approx. Ma�k R-CNN ground truth Real image Real image Our Our results method Ground Ground truth truth 34

Po Poster #60 E ast Exhibition Hall B + C 35

Emer Emergenc ence o e of O f Obj bjec ect S Seg egmen - PowerPoint PPT Presentation

Emer Emergenc ence o e of O f Obj bjec ect S Seg egmen mentatio ion n in P in Per ertur urbed Gener bed Generativ ive M e Models dels Adam Bielski, Paolo Favaro University of Bern Ob Object se segm gmentation Segmentation

Odor Odor Contr Control, ol, Cor Corrosion Contr osion Control, ol, & Emer &

Su Sust stai aina nabi bility lity for r a N a Neu euro- Em Emer ergen ent t Tel

OF 23 23 A PRIL IL 200 2008 C HA HAL M ARTI ELEC ECTED EGMEN ENTS TS O TIN 1 P RES TO M OUNT

OVE VERVI VIEW EW Emer ergen gency cy and T d Trauma Care S e System ems The

Why Recovery? under a state of emergenc y 900,000 Hectares 262 Burnt Wildfires

Prof. Emer. Dr. Prasert Prof. Emer. Dr. Prasert Thongcharoen Thongcharoen Fellow of the Royal

Getting M g Mor ore P e Per erformance w ce with Pol olym ymorphism sm f from om E Emer

Dec Declaring a a Climate E e Emer ergen ency in Minnea eapol olis City of Minneapolis

Novel No el and E and Emer ergi ging t ng tar arge gets: : University of California

Calf Rearing and Calf Health Emer Kennedy Animal & Grassland Research and Innovation Centre,

Pr Prom omot oting ing Gr Gree een n Job obs: s: Experi perience ence in In India dia

Maintaining E Excel ellen ence i e in Maintaining E Excel ellen ence i e in Columbia P

Japan's orld ar 1 Emer in ensions S immi ration 1905:

ED EDA A Defence ence Pr Procureme curement nt Ga Gate teway ED EDA Defence ence

Emer ergency S Situation ons a and Crisi sis s Plans s for or P PEG C Chan annels CITY

Sha hapi ping ng B Beh ehav avior or with th Em Emer ergi ging ng Rea eal Ti Time e

Computer Vision Lecture 5: Edges, binary images and blobs Last lecture Convolution masks as

Beyond Detection: Towards Multi-Object Tracking and Segmentation Andreas Geiger Autonomous

Image restoration IMAGE P ROCES S IN G IN P YTH ON Rebeca Gonzalez Data Engineer Restore an

Energy for Business Innovate & Collaborate with the University of Nottingham Dr. Laura

Masking TablesAn Underestimated Security Risk Michael Tunstall Carolyn Whitnall Elisabeth

Masking against Side-Channel Attacks: a Formal Security Proof Matthieu Rivain Joint work with

Travel Sentiment Study Wave 16 JULY 21, 2020 COVID-19 TRAVEL SENTIMENT STUDY WAVE 16 Fielded

1. procedure ONE TO ALL BC( d , my id , X ) 2. begin mask := 2 d 1; 3. /* Set all d bits of

Emer Emergenc ence o e of O f Obj bjec ect S Seg egmen - PowerPoint PPT Presentation

Emer Emergenc ence o e of O f Obj bjec ect S Seg egmen mentatio ion n in P in Per ertur urbed Gener bed Generativ ive M e Models dels Adam Bielski, Paolo Favaro University of Bern Ob Object se segm gmentation Segmentation

Odor Odor Contr Control, ol, Cor Corrosion Contr osion Control, ol, &amp; Emer &amp;

Su Sust stai aina nabi bility lity for r a N a Neu euro- Em Emer ergen ent t Tel

OF 23 23 A PRIL IL 200 2008 C HA HAL M ARTI ELEC ECTED EGMEN ENTS TS O TIN 1 P RES TO M OUNT

OVE VERVI VIEW EW Emer ergen gency cy and T d Trauma Care S e System ems The

Why Recovery? under a state of emergenc y 900,000 Hectares 262 Burnt Wildfires

Prof. Emer. Dr. Prasert Prof. Emer. Dr. Prasert Thongcharoen Thongcharoen Fellow of the Royal

Getting M g Mor ore P e Per erformance w ce with Pol olym ymorphism sm f from om E Emer

Dec Declaring a a Climate E e Emer ergen ency in Minnea eapol olis City of Minneapolis

Novel No el and E and Emer ergi ging t ng tar arge gets: : University of California

Calf Rearing and Calf Health Emer Kennedy Animal &amp; Grassland Research and Innovation Centre,

Pr Prom omot oting ing Gr Gree een n Job obs: s: Experi perience ence in In India dia

Maintaining E Excel ellen ence i e in Maintaining E Excel ellen ence i e in Columbia P

Japan's orld ar 1 Emer in ensions S immi ration 1905:

ED EDA A Defence ence Pr Procureme curement nt Ga Gate teway ED EDA Defence ence

Emer ergency S Situation ons a and Crisi sis s Plans s for or P PEG C Chan annels CITY

Sha hapi ping ng B Beh ehav avior or with th Em Emer ergi ging ng Rea eal Ti Time e

Computer Vision Lecture 5: Edges, binary images and blobs Last lecture Convolution masks as

Beyond Detection: Towards Multi-Object Tracking and Segmentation Andreas Geiger Autonomous

Image restoration IMAGE P ROCES S IN G IN P YTH ON Rebeca Gonzalez Data Engineer Restore an

Energy for Business Innovate &amp; Collaborate with the University of Nottingham Dr. Laura

Masking TablesAn Underestimated Security Risk Michael Tunstall Carolyn Whitnall Elisabeth

Masking against Side-Channel Attacks: a Formal Security Proof Matthieu Rivain Joint work with

Travel Sentiment Study Wave 16 JULY 21, 2020 COVID-19 TRAVEL SENTIMENT STUDY WAVE 16 Fielded

1. procedure ONE TO ALL BC( d , my id , X ) 2. begin mask := 2 d 1; 3. /* Set all d bits of

Odor Odor Contr Control, ol, Cor Corrosion Contr osion Control, ol, & Emer &

Calf Rearing and Calf Health Emer Kennedy Animal & Grassland Research and Innovation Centre,

Energy for Business Innovate & Collaborate with the University of Nottingham Dr. Laura