Reassessing the Limitations of CNN Methods for Camera Pose Regression

Ng, Tony; Lopez-Rodriguez, Adrian; Balntas, Vassileios; Mikolajczyk, Krystian

doi:10.48550/arxiv.2108.07260

Cited by 4 publications

(5 citation statements)

References 51 publications

(134 reference statements)

Supporting

Mentioning

Contrasting

Order By: Relevance

“…CNN use absolute camera pose regression (APR) techniques to predict the camera pose of an input image [118] by implicitly expressing a scene using network weights. They adhere to the same process: a base network extracts the absolute features [119] that are then embedded in a high-dimensional space.…”

Section: ) Absolute Camera Pose Regressionmentioning

confidence: 99%

A Review of Recurrent Neural Network Based Camera Localization for Indoor Environments

et al. 2023

View full text Add to dashboard Cite

Camera localization involves the estimation of the camera pose of an image from a random scene. We used a single image or sequence of images or videos as the input. The output depends on the representation of the scene and method used. Several computer vision applications, such as robot navigation and safety inspection, can benefit from camera localization. Camera localization is used to determine the position of an object on the camera in an image containing multiple images in a sequence. Structure-based localization techniques have achieved considerable success owing to a combination of image matching and coordinate regression. Absolute and relative pose regression techniques can provide end-to-end learning; however, they exhibit poor accuracies. Despite the rapid growth in computer vision, there has been no thorough review of the categorization, evaluation, and synthesis of structures and regression-based techniques. Input format and loss strategies for recurrent neural networks (RNN) have not been adequately described in the literature. The main topic is indoor camera pose regression, which is a part of the camera localization techniques. First, we discuss certain application areas for camera localization. We then discuss different camera localization techniques, such as feature and structure-based, absolute and relative pose regression techniques, and simultaneous localization and mapping (SLAM). We evaluated the frequently used datasets and qualitatively compared the absolute and relative camera pose estimation approaches. Finally, we discuss potential directions for future research, such as optimizing the computational cost of the features and evaluating the end-to-end characteristics of multiple cameras.INDEX TERMS Camera pose regression, absolute pose regression, indoor positioning, camera localization, robot navigation, SLAM.

show abstract

Section: ) Absolute Camera Pose Regressionmentioning

confidence: 99%

A Review of Recurrent Neural Network Based Camera Localization for Indoor Environments

et al. 2023

View full text Add to dashboard Cite

show abstract

“…Naseer and Burgard [26] showed that RGB-D data can be exploited to generate additional views from the limited training images to improve performance. Recently, Ng et al [27] and Moreau et al [15] extended this idea to RGB data. Other works propose to use additional information often available in robotic applications [11], [13].…”

Section: Related Workmentioning

confidence: 99%

A Probabilistic Framework for Visual Localization in Ambiguous Scenes

Zangeneh¹,

Bruns²,

Dekel³

et al. 2023

Preprint

View full text Add to dashboard Cite

Visual localization allows autonomous robots to relocalize when losing track of their pose by matching their current observation with past ones. However, ambiguous scenes pose a challenge for such systems, as repetitive structures can be viewed from many distinct, equally likely camera poses, which means it is not sufficient to produce a single best pose hypothesis. In this work, we propose a probabilistic framework that for a given image predicts the arbitrarily shaped posterior distribution of its camera pose. We do this via a novel formulation of camera pose regression using variational inference, which allows sampling from the predicted distribution. Our method outperforms existing methods on localization in ambiguous scenes. Code and data will be released at github.com/efreidun/vapor.

show abstract

“…Camera pose prediction typically involves estimating the relationship between two consecutive datasets with six degrees of freedom (DoF) [76][77][78][79]. The primary 2D-planebased loss function, as discussed earlier, acts as the central mechanism for training the camera pose prediction algorithm, while the extended loss function indirectly addresses occlusions resulting from obscured objects.…”

Section: Self-supervised Loss Functions Including Depth Consistencymentioning

confidence: 99%

“…In the field of unsupervised monocular depth prediction, incorporating a camera pose network is crucial for determining the relative positions of cameras [7][8][9][10][11]. A significant challenge in this area has been achieving accurate six degrees of freedom (DoF) camera pose predictions [76][77][78][79]. Two principal methodologies have been explored: one method inputs moving objects and depth data directly, while the other utilizes a visual odometry algorithm based on a distinct mathematical function [9].…”

Section: Camera Pose Estimationmentioning

confidence: 99%

Switchable-Encoder-Based Self-Supervised Learning Framework for Monocular Depth and Pose Estimation

Kim,

Gao,

Park

et al. 2023

Remote Sensing

View full text Add to dashboard Cite

Monocular depth prediction research is essential for expanding meaning from 2D to 3D. Recent studies have focused on the application of a newly proposed encoder; however, the development within the self-supervised learning framework remains unexplored, an aspect critical for advancing foundational models of 3D semantic interpretation. Addressing the dynamic nature of encoder-based research, especially in performance evaluations for feature extraction and pre-trained models, this research proposes the switchable encoder learning framework (SELF). SELF enhances versatility by enabling the seamless integration of diverse encoders in a self-supervised learning context for depth prediction. This integration is realized through the direct transfer of feature information from the encoder and by standardizing the input structure of the decoder to accommodate various encoder architectures. Furthermore, the framework is extended and incorporated into an adaptable decoder for depth prediction and camera pose learning, employing standard loss functions. Comparative experiments with previous frameworks using the same encoder reveal that SELF achieves a 7% reduction in parameters while enhancing performance. Remarkably, substituting newly proposed algorithms in place of an encoder improves the outcomes as well as significantly decreases the number of parameters by 23%. The experimental findings highlight the ability of SELF to broaden depth factors, such as depth consistency. This framework facilitates the objective selection of algorithms as a backbone for extended research in monocular depth prediction.

show abstract

Reassessing the Limitations of CNN Methods for Camera Pose Regression

Cited by 4 publications

References 51 publications

A Review of Recurrent Neural Network Based Camera Localization for Indoor Environments

A Review of Recurrent Neural Network Based Camera Localization for Indoor Environments

A Probabilistic Framework for Visual Localization in Ambiguous Scenes

Switchable-Encoder-Based Self-Supervised Learning Framework for Monocular Depth and Pose Estimation

Contact Info

Product

Resources

About