Download presentation
Presentation is loading. Please wait.
Published byFarida Yuwono Modified over 5 years ago
1
Learning Compositional Visual Concepts with Mutual Consistency
Yunye Gong, Srikrishna Karanam, Ziyan Wu, Kuan-Chuan Peng, Jan Ernst, and Peter C. Doerschuk CVPR 2018 (spotlight) Concept GAN Presented by: Anand Bhattad (Vision Lunch, April )
2
Image Translation (Pix to Pix and Cycle GAN)
Pix2Pix Cycle GAN Source: Unpaired Image-to-Image Translation using Cycle-Consistent Adversarial
3
Cycle GAN
4
Source: Unpaired Image-to-Image Translation using Cycle-Consistent Adversarial
5
New Unseen Samples from Translations?
Colored Bags Edge Bags Colored Shoes Edge Shoes
6
Overview: Cycle GAN G1 F1 F2 G2 F2 G2 G1 F1
Cycle GAN: Learn and Compose concepts separately G1 F1 G2 F2 G1 F1 G2 F2
7
Losses: Adversarial G1 F1 G2 F2
8
Losses: Cycle Consistency
9
Can we generate sketch shoes by combining them?
F1 F2 G2 F2 G2 G1 F1
10
Proposed Model: Concept GAN
Cycle GAN: Learn and Compose concepts separately Concept GAN: Learn and Compose concepts simultaneously G1 F1 F2 F2 G2 F2 G2 G1 F1
11
Losses: Cycle Consistency
12
Losses: Cycle Consistency (Dist=4)
G1 G1 F1 F1 F2 G2 F2 G2
13
Losses: Cycle Consistency (Dist=4)
G1 G1 F1 F1 F2 G2 F2 G2
14
Losses: Cycle Consistency (Dist=4)
G1 G1 F1 F1 F2 G2 F2 G2
15
Loss: Commutative G1 G1 F1 F1 F2 G2 F2 G2
16
Loss: Commutative G1 F1 F2 G2 F2 G2 G1 F1
17
Loss: Commutative G1 F1 F2 G2 F2 G2 G1 F1
18
Concept GAN
19
Comparing Results
20
Qualitative Results
21
Qualitative Results
22
Quantitative Results (Classification)
Test Set: edge Shoes only Test Set: with eyeglasses, with bangs 18 60 (Accuracy if L_x removed)
23
Ablation Study
24
Re-ID
25
Generalizing to n(=3) Concepts
26
Learning from 2 attributes in 2 experiments
Experiment1: (smile, eyeglasses) Experiment2: (bangs, eyeglasses) (smile, eyeglasses, bangs)
27
Yaxing Wang, Joost van de Weijer, Luis Herranz CVPR 2018
Mix and match networks: encoder-decoder alignment for zero-pair image translation Yaxing Wang, Joost van de Weijer, Luis Herranz CVPR 2018
28
Task Mix and match networks do not require retraining on unseen transformations, in contrast to unpaired translation alternatives (e.g. CycleGAN) leverages depth and semantic segmentation, ignoring the available RGB data and the corresponding pairs
29
Overview
30
Losses
31
Depth –to- Semantic segmentation
32
Depth-to-Semantic segmentation
Similar presentations
© 2024 SlidePlayer.com. Inc.
All rights reserved.