ENG: End-to-end Neural Geometry for Robust Depth and Pose Estimation using CNNs

Dharmasiri, Thanuja; Spek, Andrew; Drummond, Tom

doi:10.48550/arxiv.1807.05705

Cited by 5 publications

(11 citation statements)

References 0 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Increasing the sequence length beyond a sequence length of 5 examines its ability to generalise beyond the length of the training sequences. Both DeepVO [39] and ENG [11] lack an internal map representation to localise against and similarly both methods suffer from a larger accumulated drift towards the end of the sequence. MapNet [19] fairs better, although suffers from inaccuracies due to cell quantisation and false pose modalities which appear over longer periods of time.…”

Section: Quantitative Resultsmentioning

confidence: 99%

“…For the Doom dataset, we compared our framework against DeepVO [39] and MapNet [19], both which maintain an internal representation of previously seen observations. Additionally, we compared our models against a recent state-of-the-art frame-to-frame approach ENG [11]. For the AVD dataset, we also compare with a mature classic SLAM baseline, an RGB-D implementation of ORB-SLAM2 [29].…”

Section: Methodsmentioning

confidence: 99%

“…Frame-based Prior to the introduction of memory based models for localisation and mapping tasks, frame-by-frame methods [8,23] and more recently [11,28], explored the exploitation of geometric constraints for reducing the search space when optimising Convolutional Neural Networks (CNN). The pioneering work of [23] applied direct pose regression for inferring the relative pose between two views.…”

Section: Related Workmentioning

confidence: 99%

“…The work by [8] enhanced the information provided to the regression network by including the optical flow between two consecutive frames. A natural extension was explored by [11] which simultaneously estimated a depth map along with a latent optical flow constraint for regressing the pose between consecutive frames. CodeSLAM [3] optimises an encoder-decoder setup to efficiently represent depth frames as latent codes.…”

Section: Related Workmentioning

confidence: 99%

“…In future work, we look towards extending EMP-Net to larger navigation problems by addressing the linear com- ORB-SLAM2 (RGB-D) [29] 432 3090 794 DeepVO [39] 220.0 1690 741 MapNet [19] 312.3 1680 601 ENG [11] 234. plexity growth of computing the correspondence matrix (i.e. large buffer sizes).…”

Section: Future Workmentioning

confidence: 99%

See 4 more Smart Citations

EMPNet: Neural Localisation and Mapping Using Embedded Memory Points

Avraham¹,

Zuo²,

Dharmasiri³

et al. 2019

Preprint

Self Cite

View full text Add to dashboard Cite

show abstract

Section: Quantitative Resultsmentioning

confidence: 99%

Section: Methodsmentioning

confidence: 99%

Section: Related Workmentioning

confidence: 99%

Section: Related Workmentioning

confidence: 99%

Section: Future Workmentioning

confidence: 99%

See 3 more Smart Citations

EMPNet: Neural Localisation and Mapping Using Embedded Memory Points

Avraham¹,

Zuo²,

Dharmasiri³

et al. 2019

Preprint

Self Cite

View full text Add to dashboard Cite

show abstract

Look No Deeper: Recognizing Places from Opposing Viewpoints under Varying Scene Appearance using Single-View Depth Estimation

Garg

Babu

Dharmasiri

et al. 2019

2019 International Conference on Robotics and Automation (ICRA)

Self Cite

View full text Add to dashboard Cite

Visual place recognition (VPR) -the act of recognizing a familiar visual place -becomes difficult when there is extreme environmental appearance change or viewpoint change. Particularly challenging is the scenario where both phenomena occur simultaneously, such as when returning for the first time along a road at night that was previously traversed during the day in the opposite direction. While such problems can be solved with panoramic sensors, humans solve this problem regularly with limited field of view vision and without needing to constantly turn around. In this paper, we present a new depth-and temporal-aware visual place recognition system that solves the opposing viewpoint, extreme appearance-change visual place recognition problem. Our system performs sequence-to-single matching by extracting depth-filtered keypoints using a stateof-the-art depth estimation pipeline, constructing a keypoint sequence over multiple frames from the reference dataset, and comparing those keypoints to those in a single query image. We evaluate the system on a challenging benchmark dataset and show that it consistently outperforms state-of-theart techniques. We also develop a range of diagnostic simulation experiments that characterize the contribution of depth-filtered keypoint sequences with respect to key domain parameters including degree of appearance change and camera motion.

show abstract

CReaM: Condensed Real-time Models for Depth Prediction using Convolutional Neural Networks

Spek

Dharmasiri

Drummond

2018

2018 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS)

Self Cite

View full text Add to dashboard Cite

Since the resurgence of CNNs the robotic vision community has developed a range of algorithms that perform classification, semantic segmentation and structure prediction (depths, normals, surface curvature) using neural networks. While some of these models achieve state-of-the art results and super human level performance, deploying these models in a time critical robotic environment remains an ongoing challenge. Real-time frameworks are of paramount importance to build a robotic society where humans and robots integrate seamlessly. To this end, we present a novel real-time structure prediction framework that predicts depth at 30 frames per second on an NVIDIA-TX2. At the time of writing, this is the first piece of work to showcase such a capability on a mobile platform. We also demonstrate with extensive experiments that neural networks with very large model capacities can be leveraged in order to train accurate condensed model architectures in a "from teacher to student" style knowledge transfer.

show abstract

ENG: End-to-end Neural Geometry for Robust Depth and Pose Estimation using CNNs

Cited by 5 publications

References 0 publications

EMPNet: Neural Localisation and Mapping Using Embedded Memory Points

EMPNet: Neural Localisation and Mapping Using Embedded Memory Points

Look No Deeper: Recognizing Places from Opposing Viewpoints under Varying Scene Appearance using Single-View Depth Estimation

CReaM: Condensed Real-time Models for Depth Prediction using Convolutional Neural Networks

Contact Info

Product

Resources

About