Recurrent Dilated DenseNets for a Time-Series Segmentation Task

Fuchs, Alexander; Priewald, Robin; Pernkopf, Franz

doi:10.1109/icmla.2019.00021

Cited by 7 publications

(4 citation statements)

References 11 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…However, we remark that these methods Schuster et al, 2019;Takahashi and Mitsufuji, 2021] simply use convolution with different dilation rates to compute the multiscale feature descriptor, while the fine-grained features could be over-smoothed and become indistinguishable after subsampling. This prominent problem is also known as aliasing [Gong and Poellabauer, 2018], which degrades the performance of CNN-based recognition tasks Fuchs et al, 2019]. Inspired by an interesting property of recursivity principle in scale-space [Pauwels et al, 1995], this paper presents recursive Hermite polynomials to prevent the occurrence of aliasing in fine-grained features.…”

Section: Multi-scale Representation Learningmentioning

confidence: 99%

See 1 more Smart Citation

Multi-scale Spatial Representation Learning via Recursive Hermite Polynomial Networks

Feng

Tian

Zhang

et al. 2022

Proceedings of the Thirty-First International Joint Conference on Artificial Intelligence

View full text Add to dashboard Cite

Semi-Supervised Object Detection (SSOD) aims to improve performance by leveraging a large amount of unlabeled data. Existing works usually adopt the teacher-student framework to enforce student to learn consistent predictions over the pseudo-labels generated by teacher. However, the performance of the student model is limited since the noise inherently exists in pseudo-labels. In this paper, we investigate the causes and effects of noisy pseudo-labels and propose a simple yet effective approach denoted as Self-Correction Mean Teacher(SCMT) to reduce the adverse effects. Specifically, we propose to dynamically re-weight the unsupervised loss of each student's proposal with additional supervision information from the teacher model, and assign smaller loss weights to possible noisy proposals. Extensive experiments on MS-COCO benchmark have shown the superiority of our proposed SCMT, which can significantly improve the supervised baseline by more than 11% mAP under all 1%, 5% and 10% COCO-standard settings, and surpasses state-of-the-art methods by about 1.5% mAP. Even under the challenging COCO-additional setting, SCMT still improves the supervised baseline by 4.9% mAP, and significantly outperforms previous methods by 1.2% mAP, achieving a new state-of-the-art performance.

show abstract

Section: Multi-scale Representation Learningmentioning

confidence: 99%

“…This is especially obvious for the fine-grained features with much higher frequency. Thus, an appropriate lowpass filter for anti-artifacts (e.g., a standard convolution filter Fuchs et al, 2019]) is needed to boost the accuracy of dense prediction tasks.…”

Section: Introductionmentioning

confidence: 99%

Multi-scale Spatial Representation Learning via Recursive Hermite Polynomial Networks

Feng

Tian

Zhang

et al. 2022

Proceedings of the Thirty-First International Joint Conference on Artificial Intelligence

View full text Add to dashboard Cite

show abstract

“…To avoid this problem, most CNN architectures that involve dilated convolution are carefully designed to allow earlier layers to learn appropriate anti-aliasing filter if necessary, i.e., standard convolutions are applied before dilated convolutions with fixed dilation factor [1,45], or the dilation factors is gradually increased as the layer goes deeper [46,43]. A naive combination of DenseNet with dilation has already been proposed [10], where dilated convolutions are used and the dilation factor was set to one at the initial layer and doubled as the layer goes deeper. However, this approach has significant aliasing due to skip connections, as discussed in Sec.…”

Section: Dilated Convolution and Aliasingmentioning

confidence: 99%

Densely connected multidilated convolutional networks for dense prediction tasks

Mitsufuji

2020

Preprint

View full text Add to dashboard Cite

Tasks that involve high-resolution dense prediction require a modeling of both local and global patterns in a large input field. Although the local and global structures often depend on each other and their simultaneous modeling is important, many convolutional neural network (CNN)based approaches interchange representations in different resolutions only a few times. In this paper, we claim the importance of a dense simultaneous modeling of multiresolution representation and propose a novel CNN architecture called densely connected multidilated DenseNet (D3Net). D3Net involves a novel multidilated convolution that has different dilation factors in a single layer to model different resolutions simultaneously. By combining the multidilated convolution with the DenseNet architecture, D3Net incorporates multiresolution learning with an exponentially growing receptive field in almost all layers, while avoiding the aliasing problem that occurs when we naively incorporate the dilated convolution in DenseNet. Experiments on the image semantic segmentation task using Cityscapes and the audio source separation task using MUSDB18 show that the proposed method has superior performance over stateof-the-art methods.

show abstract

“…It is inefficient to have many parameters, especially for high-resolution data, especially when local features are converted to global ones. In [21,22], a network design was proposed that combines the advantages of DenseNet with the advantages of dilated convolution. In fact, typical dilated convolutions were used, and the dilatation factors were calculated based on the layer depth, resulting in significant aliasing effects.…”

Section: Introductionmentioning

confidence: 99%

Single channel speech enhancement using time-frequency attention mechanism based nested U-net model

Prathipati,

Chakravarthy

2024

Eng. Res. Express

View full text Add to dashboard Cite

Deep-learning models have used attention mechanisms to improve the quality and intelligibility of noisy speech, demonstrating the effectiveness of attention mechanisms. We rely on either spatial or temporal-based attention mechanisms, resulting in severe information loss. In this paper, a time-frequency attention mechanism with a nested U-network (TFANUNet) is proposed for single-channel speech enhancement. By using time-frequency attention (TFA), learns the channel, frequency and time information which is more significant for speech enhancement. Basically, the proposed model is an encoder-decoder model, where each layer in the encoder and decoder is followed by a nested dense residual dilated DensNet (NDRD) based multi-scale context aggression block. NDRD involves multiple dilated convolution with different dilatation factors to explore the large receptive area at different scales simultaneously. NDRD avoids the aliasing problem in DenseNet. We integrated the TFA and NDRD blocks into the proposed model to enable refined feature set extraction without information loss and utterance-level context aggregation, respectively. The proposed TFANUNet model results outperform baselines in terms of STOI and PESQ.

show abstract

Recurrent Dilated DenseNets for a Time-Series Segmentation Task

Cited by 7 publications

References 11 publications

Multi-scale Spatial Representation Learning via Recursive Hermite Polynomial Networks

Multi-scale Spatial Representation Learning via Recursive Hermite Polynomial Networks

Densely connected multidilated convolutional networks for dense prediction tasks

Single channel speech enhancement using time-frequency attention mechanism based nested U-net model

Contact Info

Product

Resources

About