AdaDepth: Unsupervised Content Congruent Adaptation for Depth Estimation

Kundu, Jogendra Nath; Uppala, Phani Krishna; Pahuja, Anuj; Babu, R. Venkatesh

doi:10.1109/cvpr.2018.00281

Cited by 150 publications

(51 citation statements)

References 31 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In [24], and the ones with the unsupervised setting, i.e. Zhou et al [29], AdaDepth [9], Garg et al [4] and Godard et al [6]. Among all the supervised approaches, we have achieved very competitive performance to the best one of them (i.e.…”

Section: State Of the Art Comparisonmentioning

confidence: 88%

“…There are few works in the literature considering GAN models for the more challenging depth estimation task. Although Kundu et al [9] investigate adversarial learning for the task, they utilize it in a context of domain adaptation in a single-track network, using a semi-supervised setting with an extra synthetic dataset, while ours considers a fully unsupervised setting and the adversarial learning in a cycled generative network aims to help the reconstruction of better image views. Both the intuition and the network design are significantly different.…”

Section: Unsupervised Depth Estimationmentioning

confidence: 99%

“…Godard et al [6]). AdaDepth [9] is the most technically related to our approach, which considers adversarial learning in a context of domain adaptation with extra synthetic training data. Ours significantly outperforms their results with both the supervised and unsupervised setting, further demonstrating the effectiveness of the means we considered and proposed for unsupervised depth estimation with the adversarial learning strategy.…”

Section: State Of the Art Comparisonmentioning

confidence: 99%

See 2 more Smart Citations

Unsupervised Adversarial Depth Estimation Using Cycled Generative Networks

Pilzer

Puscas

et al. 2018

2018 International Conference on 3D Vision (3DV)

149

126

View full text Add to dashboard Cite

While recent deep monocular depth estimation approaches based on supervised regression have achieved remarkable performance, costly ground truth annotations are required during training. To cope with this issue, in this paper we present a novel unsupervised deep learning approach for predicting depth maps and show that the depth estimation task can be effectively tackled within an adversarial learning framework. Specifically, we propose a deep generative network that learns to predict the correspondence field (i.e. the disparity map) between two image views in a calibrated stereo camera setting. The proposed architecture consists of two generative subnetworks jointly trained with adversarial learning for reconstructing the disparity map and organized in a cycle such as to provide mutual constraints and supervision to each other. Extensive experiments on the publicly available datasets KITTI and Cityscapes demonstrate the effectiveness of the proposed model and competitive results with state of the art methods. The code and trained model are available: https://github.com/ andrea-pilzer/unsup-stereo-depthGAN

show abstract

Section: State Of the Art Comparisonmentioning

confidence: 88%

Section: Unsupervised Depth Estimationmentioning

confidence: 99%

Section: State Of the Art Comparisonmentioning

confidence: 99%

See 1 more Smart Citation

Unsupervised Adversarial Depth Estimation Using Cycled Generative Networks

Pilzer

Puscas

et al. 2018

2018 International Conference on 3D Vision (3DV)

149

126

View full text Add to dashboard Cite

show abstract

“…synthetic data to pre-train the networks and fine-tune on real data. For tasks like optical flow [24] and depth estimation [15,19,10] synthetic data is much more accessible and does not contain the various artifacts and sampling noises that real data has. However, this approach suffers from the domain shift problem where the data distribution of synthetic and real data can differ.…”

Section: Related Workmentioning

confidence: 99%

Unsupervised Batch Normalization

Kocyigit

Sevilla-Lara

Hospedales

et al. 2020

2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW)

View full text Add to dashboard Cite

Batch Normalization is a widely used tool in neural networks to improve the generalization and convergence of training. However, on small datasets due to the difficulty of obtaining unbiased batch statistics it cannot be applied effectively. In some cases, even if there is only a small labeled dataset available, there are larger unlabeled datasets from the same distribution. We propose using such unlabeled examples to calculate batch normalization statistics, which we call Unsupervised Batch Normalization (UBN). We show that using unlabeled examples for batch statistic calculations results in a reduction of the bias of the statistics, as well as regularization leveraging the data manifold. UBN is easy to implement, computationally inexpensive and can be applied to a variety problems. We report results on monocular depth estimation, where obtaining dense labeled examples is difficult and expensive. Using unlabeled samples, and UBN, we obtain an increase in accuracy of more than 6% on the KITTI dataset, compared to using traditional batch normalization only on the labeled samples.

show abstract

“…This property makes adversarial learning suitable for unsupervised depth estimation, allowing the generation of more accurate viewpoint images. Pilzer et al [20] propose to learn a stereo matching model in a cycled, adversarial fashion, while [13] utilize it in a context of domain adaptation for a single-track network, using a semi-supervised setting with additional synthetic data. In this work, we use adversarial learning at the generator level, however, we present a novel dual GAN design and use a CRF model to couple the network for a structured refinement and fusion of both generator and discriminator outputs in an end-to-end fashion.…”

Section: Related Workmentioning

confidence: 99%

Structured Coupled Generative Adversarial Networks for Unsupervised Monocular Depth Estimation

Puscas¹,

Pilzer

et al. 2019

2019 International Conference on 3D Vision (3DV)

View full text Add to dashboard Cite

Inspired by the success of adversarial learning, we propose a new end-to-end unsupervised deep learning framework for monocular depth estimation consisting of two Generative Adversarial Networks (GAN), deeply coupled with a structured Conditional Random Field (CRF) model. The two GANs aim at generating distinct and complementary disparity maps and at improving the generation quality via exploiting the adversarial learning strategy. The deep CRF coupling model is proposed to fuse the generative and discriminative outputs from the dual GAN nets. As such, the model implicitly constructs mutual constraints on the two network branches and between the generator and discriminator. This facilitates the optimization of the whole network for better disparity generation. Extensive experiments on the KITTI, Cityscapes, and Make3D datasets clearly demonstrate the effectiveness of the proposed approach and show superior performance compared to state of the art methods. The code and models are available at https://github.com/mihaipuscas/ 3dv---coupled-crf-disparity. * Equal contribution. † Work performed while at SAP ML Research Berlin CRF Structured Coupling of G and D Structured Coupling of Two BranchesG a < l a t e x i t s h a 1 _ b a s e 6 4 = " N V N n C l + g V N 2 m M C u n S 5 W c s 8 h s UY H w z 8 9 t P T G k e y 0 c z S Z g f 4 V D y k F M 0 V n q 4 7 s Z l k h o m 6 W J R m A p i Y j L 7 m w y 4 Y t S I i S V I F b e 3 E j p C h d T Y d E o 2 B G / 5 5 V X S u q h 6 b t W 7 v 6 z U r / M 4 i n A C p 3 A O H t S g D n f Q g C Z Q G M I z v M K b I 5 w X 5 9 3 5 W L Q W n H z m G P 7 A + f w B B / y N m Q = = < / l a t e x i t > < l a t e x i t s h a 1 _ b a s e 6 4 = " N V N n C l + g V N 2 m M C u n S 5 W c s 8 h s U O w = " > A A A B 6 n i c b V A 9 S w N B E J 2 L X z F + R S 1 t F o N g F e 5 E i I V F w E L L i O Y D k i P M b f a S J X t 7 x + 6 e E I 7 8 B B s L R W z 9 R X b + G z f J F Z r 4 Y O D x 3 g w z 8 4 J E c G 1 c 9 9 s p r K 1 v b G 4 V t 0 s 7 u 3 v 7 B + X D o 5 a O U 0 V Z k 8 Y i V p 0 A N R N c s q b h R r B O o h h G g W D t Y H w z 8 9 t P T G k e y 0 c z S Z g f 4 V D y k F M 0 V n q 4 7 s Z l k h o m 6 W J R m A p i Y j L 7 m w y 4 Y t S I i S V I F b e 3 E j p C h d T Y d E o 2 B G / 5 5 V X S u q h 6 b t W 7 v 6 z U r / M 4 i n A C p 3 A O H t S g D n f Q g C Z Q G M I z v M K b I 5 w X 5 9 3 5 W L Q W n H z m G P 7 A + f w B B / y N m Q = = < / l a t e x i t > < l a t e x i t s h a 1 _ b a s e 6 4 = " N V N n C l + g V N 2 m M C u n S 5 W c s 8 h s U O w = " > A A A B 6 n i c b V A 9 S w N B E J 2 L X z F + R S 1 t F o N g F e 5 E i I V F w E L L i O Y D k i P M b f a S J X t 7 x + 6 e E I 7 8 B B s L R W z 9 R X b + G z f J F Z r 4 Y O D x 3 g w z 8 4 J E c G 1 c 9 9 s p r K 1 v b G 4 V t 0 s 7 u 3 v 7 B + X D o 5 a O U 0 V Z k 8 Y i V p 0 A N R N c s q b h R r B O o h h G g W D t Y H w z 8 9 t P T G k e y 0 c z S Z g f 4 V D y k F M 0 V n q 4 7 s Z l k h o m 6 W J R m A p i Y j L 7 m w y 4 Y t S I i S V I F b e 3 E j p C h d T Y d E o 2 B G / 5 5 V X S u q h 6 b t W 7 v 6 z U r / M 4 ...

show abstract

AdaDepth: Unsupervised Content Congruent Adaptation for Depth Estimation

Cited by 150 publications

References 31 publications

Unsupervised Adversarial Depth Estimation Using Cycled Generative Networks

Unsupervised Adversarial Depth Estimation Using Cycled Generative Networks

Unsupervised Batch Normalization

Structured Coupled Generative Adversarial Networks for Unsupervised Monocular Depth Estimation

Contact Info

Product

Resources

About