RGB-D Object Recognition Using Deep Convolutional Neural Networks

Zia, Saman; Yüksel, Buket; Yüret, Deniz; Yemez, Y.

doi:10.1109/iccvw.2017.109

Cited by 39 publications

(24 citation statements)

References 17 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The problem of estimating the depth map of a scene from its single image has attracted a lot of attention in the field of computer vision, since depth maps have a lot of applications, such as augmented reality [22], human computer interaction [33], human activity recognition [12], scene recognition [41], and segmentation [27,32]. The recent employment of convolutional neural networks (CNNs) has accelerated the research of the problem [9,25,23,2,4].…”

Section: Introductionmentioning

confidence: 99%

Revisiting Single Image Depth Estimation: Toward Higher Resolution Maps With Accurate Object Boundaries

Özay

Zhang

et al. 2019

2019 IEEE Winter Conference on Applications of Computer Vision (WACV)

346

451

View full text Add to dashboard Cite

This paper considers the problem of single image depth estimation. The employment of convolutional neural networks (CNNs) has recently brought about significant advancements in the research of this problem. However, most existing methods suffer from loss of spatial resolution in the estimated depth maps; a typical symptom is distorted and blurry reconstruction of object boundaries. In this paper, toward more accurate estimation with a focus on depth maps with higher spatial resolution, we propose two improvements to existing approaches. One is about the strategy of fusing features extracted at different scales, for which we propose an improved network architecture consisting of four modules: an encoder, decoder, multi-scale feature fusion module, and refinement module. The other is about loss functions for measuring inference errors used in training. We show that three loss terms, which measure errors in depth, gradients and surface normals, respectively, contribute to improvement of accuracy in an complementary fashion. Experimental results show that these two improvements enable to attain higher accuracy than the current state-of-the-arts, which is given by finer resolution reconstruction, for example, with small objects and object boundaries.

show abstract

Section: Introductionmentioning

confidence: 99%

Revisiting Single Image Depth Estimation: Toward Higher Resolution Maps With Accurate Object Boundaries

Özay

Zhang

et al. 2019

2019 IEEE Winter Conference on Applications of Computer Vision (WACV)

346

451

View full text Add to dashboard Cite

show abstract

“…Compared deep models demonstrated relative improvements in range of 3% to 10% (see Table 1). The evaluation has been performed using our CNN-based classifier along with results published in [7], [11] and [4] acquired using several different architectures. The comparison shows that the benefit of depth modality is comparable across multiple different classification methods.…”

Section: Resultsmentioning

confidence: 99%

“…Some models, like VGG3D [11] try to utilize depth information by generating RGB voxel grids from RGB-D images. These are then fed into a modified VGGnet model which is pre-trained on a large 2D RGB dataset.…”

Section: Classification Of 3d Datamentioning

confidence: 99%

Evaluation of Depth Modality in Convolutional Neural Network Classification of RGB-D Images

Varga¹,

Jadlovský²

2019

AEI

View full text Add to dashboard Cite

This paper investigates the value of depth modality in object classification in RGB-D images. We use a simple model based on a multi-layered convolutional neural network which we train on a dataset of segmented RGB-D images of household and office objects. We evaluate and quantify the benefit of additional depth modality and its effect on classification accuracy on this dataset. Also, we compare the benefit of depth channel against the addition of color to grayscale image. Our experimental results support a conclusion, that for these categories of objects the depth modality provides a significant benefit to classification, which also outweighs the benefit of color information. Similar supporting evidence found in recent research is shown in comparison along with the resulting quantified benefit of depth modality.

show abstract

“…As an example, given all trained models, the combination of different outputs may also further improve the denoising performance. It should also be noted that Zia et al proposed to extend a 2D filter to a 3D one by replicating this 2D filter along the depth dimension for RGB-D object recognition [45]. However, that extension cannot guarantee that the initial 3D network has the same performance as the trained 2D network, i.e.…”

Section: Discussionmentioning

confidence: 99%

Correction for “3D Convolutional Encoder-Decoder Network for Low-Dose CT via Transfer Learning From a 2D Trained Network” [Jun 18 1522-1534]

Shan

Zhang

Yang

et al. 2018

IEEE Trans. Med. Imaging

View full text Add to dashboard Cite

Low-dose computed tomography (CT) has attracted major attention in the medical imaging field, since CT-associated x-ray radiation carries health risks for patients. The reduction of the CT radiation dose, however, compromises the signal-tonoise ratio, which affects image quality and diagnostic performance. Recently, deep-learning-based algorithms have achieved promising results in low-dose CT denoising, especially convolutional neural network (CNN) and generative adversarial network (GAN) architectures. This article introduces a Conveying Pathbased Convolutional Encoder-decoder (CPCE) network in 2D and 3D configurations within the GAN framework for low-dose CT denoising. A novel feature of this approach is that an initial 3D CPCE denoising model can be directly obtained by extending a trained 2D CNN, which is then fine-tuned to incorporate 3D spatial information from adjacent slices. Based on the transfer learning from 2D to 3D, the 3D network converges faster and achieves a better denoising performance when compared to a training from scratch. By comparing the CPCE network with recently published work based on the simulated Mayo dataset and the real MGH dataset, we demonstrate that the 3D CPCE denoising model has a better performance in that it suppresses image noise and preserves subtle structures.Index Terms-Low-dose CT, denoising, convolutional neural network, generative adversarial network, 3D spatial information.

show abstract

RGB-D Object Recognition Using Deep Convolutional Neural Networks

Cited by 39 publications

References 17 publications

Revisiting Single Image Depth Estimation: Toward Higher Resolution Maps With Accurate Object Boundaries

Revisiting Single Image Depth Estimation: Toward Higher Resolution Maps With Accurate Object Boundaries

Evaluation of Depth Modality in Convolutional Neural Network Classification of RGB-D Images

Correction for “3D Convolutional Encoder-Decoder Network for Low-Dose CT via Transfer Learning From a 2D Trained Network” [Jun 18 1522-1534]

Contact Info

Product

Resources

About