Monocular Depth Estimation Based on Multi-Scale Graph Convolution Networks

Fu, Junwei; Liang, Jun; Wang, Ziyang

doi:10.1109/access.2019.2961606

Cited by 10 publications

(5 citation statements)

References 34 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Imputation Method Although max pooling has been suggested as a proper way of imputation [16], we independently tested Gaussian filter and Mean pooling as possible layers for reconstructing the missing data. Mean pooling and Gaussian filter are not necessarily wind up with correct depth values.…”

Section: Ablation Studymentioning

confidence: 99%

See 1 more Smart Citation

LightDepth: A Resource Efficient Depth Estimation Approach for Dealing with Ground Truth Sparsity via Curriculum Learning

Karimi¹,

Mehrpanah²,

Rawassizadeh³

2022

Preprint

View full text Add to dashboard Cite

Advances in neural networks enable tackling complex computer vision tasks such as depth estimation of outdoor scenes at unprecedented accuracy. Promising research has been done on depth estimation. However, current efforts are computationally resource-intensive and do not consider the resource constraints of autonomous devices, such as robots and drones. In this work, we present a "fast" and "battery-efficient" approach for depth estimation. Our approach devises model-agnostic curriculumbased learning for depth estimation. Our experiments show that the accuracy of our model performs on par with the state-of-the-art models, while its response time outperforms other models by 71%. All codes are available online at https://github.com/fatemehkarimii/LightDepth.

show abstract

Section: Ablation Studymentioning

confidence: 99%

“…There are many research works for performing depth estimation on sparse data, but most of them are focused on the model architecture rather than the sparsity itself [30,16,62,12,64,44,57]. Almost all models suffer from challenges created by their ground truth sparsity.…”

mentioning

confidence: 99%

LightDepth: A Resource Efficient Depth Estimation Approach for Dealing with Ground Truth Sparsity via Curriculum Learning

Karimi¹,

Mehrpanah²,

Rawassizadeh³

2022

Preprint

View full text Add to dashboard Cite

show abstract

“…However, this causes losing many nodes of the image when 3D objects are mapped in 2D planes. Fu et al [32], created the depth topological graph from a coarse depth map and they used this graph as a depth clue in their model to avoid depth nodes losses. Although this technique generates a depth topological graph from a coarse depth map obtained from pre-trained models.…”

Section: Graph Neural Networkmentioning

confidence: 99%

GCNDepth: Self-supervised Monocular Depth Estimation based on Graph Convolutional Network

Masoumian¹,

Rashwan²,

Abdulwahab³

et al. 2021

Preprint

View full text Add to dashboard Cite

Depth estimation is a challenging task of 3D reconstruction to enhance the accuracy sensing of environment awareness. This work brings a new solution with a set of improvements, which increase the quantitative and qualitative understanding of depth maps compared to existing methods. Recently, convolutional neural network (CNN) has demonstrated its extraordinary ability in estimating depth maps from monocular videos. However, traditional CNN does not support topological structure and they can work only on regular image regions with determined size and weights. On the other hand, graph convolutional networks (GCN) can handle the convolution on non-Euclidean data and it can be applied to irregular image regions within a topological structure. Therefore, in this work in order to preserve object geometric appearances and distributions, we aim at exploiting GCN for a self-supervised depth estimation model. Our model consists of two parallel auto-encoder networks: the first is an auto-encoder that will depend on ResNet-50 and extract the feature from the input image and on multiscale GCN to estimate the depth map. In turn, the second network will be used to estimate the ego-motion vector (i.e., 3D pose) between two consecutive frames based on ResNet-18. Both the estimated 3D pose and depth map will be used for constructing a target image. A combination of loss functions related to photometric, reprojection and smoothness is used to cope with bad depth prediction and preserve the discontinuities of the objects. In particular, our method provided comparable and promising results with a high prediction accuracy of 89% on the publicly KITTI and Make3D datasets along with a reduction of 40% in the number of trainable parameters compared to the state of the art solutions. The source code is publicly available at https://github.com/ArminMasoumian/GCNDepth.git

show abstract

“…Many studies show that the multi-scale estimation theory can effectively improve the estimation accuracy of the system when compared with the traditional estimation theory. [26][27][28] Therefore, many researchers have presented a series of asynchronous information fusion algorithms for multi-sensor integrated navigation systems, basing on the theory of multi-scale estimation. These studies have made good contributions to the development of asynchronous information fusion technology of integrated navigation system.…”

Section: Introductionmentioning

confidence: 99%

“…Many studies show that the multi-scale estimation theory can effectively improve the estimation accuracy of the system when compared with the traditional estimation theory. 26–28…”

Section: Introductionmentioning

confidence: 99%

Multi-scale asynchronous fusion algorithm for multi-sensor integrated navigation system

Xueyuan

Pan²,

Sun³

et al. 2022

Proceedings of the Institution of Mechanical Engineers, Part I:

View full text Add to dashboard Cite

The asynchronous fusion method is the key technology of multi-sensor integrated navigation system which is independently observed by multiple auxiliary navigation sensors with different data output rates. The multi-scale asynchronous fusion can effectively increase the accuracy of integrated navigation system, but the existing multi-scale asynchronous fusion algorithm has the characteristic of heavy computational burden and requires the data output rates of auxiliary navigation sensors to meet unrealistic special requirements. To solve this problem, a multi-scale asynchronous fusion algorithm for the multi-sensor integrated navigation system is presented based on state block vector and wavelet transform. This algorithm first converts the system’s original state equation into the relationship between the state block vector and the state point vector to obtain the state equation of the multi-scale dynamic system, and expresses the original measurement equation as the relationship with the state block vector to obtain the measurement equation of the multi-scale dynamic system. Then, the special implementation process of this algorithm is studied and expressed in details, on the basis of the matrix operator with scale and wavelet property. Finally, a multi-sensor integrated navigation system composed of strapdown inertial navigation system, celestial navigation system, global navigation satellite system, and air data system is designed and experimented validation analysis. Experiment results show that this algorithm can significantly reduce the average root mean square error and improve the accuracy of navigation parameters, compared with asynchronous preprocessing Kalman filter and asynchronous direct Kalman filter. It would have a wide prospect in multi-sensor integrated navigation systems.

show abstract

Monocular Depth Estimation Based on Multi-Scale Graph Convolution Networks

Cited by 10 publications

References 34 publications

LightDepth: A Resource Efficient Depth Estimation Approach for Dealing with Ground Truth Sparsity via Curriculum Learning

LightDepth: A Resource Efficient Depth Estimation Approach for Dealing with Ground Truth Sparsity via Curriculum Learning

GCNDepth: Self-supervised Monocular Depth Estimation based on Graph Convolutional Network

Multi-scale asynchronous fusion algorithm for multi-sensor integrated navigation system

Contact Info

Product

Resources

About