Deeply-Fused Nets

By Jingdong Wang, Zhen Wei, Ting Zhang, Wenjun Zeng. 2016

103 citations recorded by the source corpus.

In this paper, we present a novel deep learning approach, deeply-fused nets. The central idea of our approach is deep fusion, i.e., combine the intermediate representations of base networks, where the fused output serves as the input of the remaining part of each base network, and perform such combinations deeply over several intermediate representations. The resulting deeply fused net enjoys several benefits. First, it is able to learn multi-scale representations as it enjoys the benefits of more base networks, which could form the same fused network, other than the initial group of base networks. Second, in our suggested fused net formed by one deep and one shallow base networks, the flows of the information from the earlier intermediate layer of the deep base network to the output and from the input to the later intermediate layer of the deep base network are both improved. Last, the deep and shallow base networks are jointly learnt and can benefit from each other. More interestingly, the essential depth of a fused net composed from a deep base network and a shallow base network is reduced because the fused net could be composed from a less deep base network, and thus training the fused net is less difficult than training the initial deep base network. Empirical results demonstrate that our approach achieves superior performance over two closely-related methods, ResNet and Highway, and competitive performance compared to the state-of-the-arts.

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

Citations in the corpus

Multi-view 3D Object Detection Network for Autonomous Driving (2016, methodology): “In this paper, we design a deep fusion approach inspired by FractalNet [13] and Deeply-Fused Net [23].”
Deep High-Resolution Representation Learning for Visual Recognition (2019, background): “The differences include: (1) Our fusion outputs four-resolution representations other than only one, and (2) our fusion modules are repeated several times which is inspired by deep fusion [129], [143], [155], [178], [184].”
Wider or Deeper: Revisiting the ResNet Model for Visual Recognition (2016, methodology): “s in their proposed networks as well. For example, the 76-layer Inception-v4 network [30] has a much shorter twenty-layer route from its input to the output. There might be differences in the details [32] between fusion by concatenation (Inception-v4) and fusion by summation (ResNets). However, the ”
Data-Driven Sparse Structure Selection for Deep Neural Networks (2017, background): “Their explanation is also evidenced by subsequent work in network designs [37, 38].”
Deep High-Resolution Representation Learning for Human Pose Estimation (2019, background): “then a reﬁnenet combines the low-to-high level features that are processed through convolutions. Our approach repeats multi-scale fusion, which is partially inspired by deep fusion and its extensions [67,73,59,80,82]. Intermediate supervision. Intermediate supervision or deep supervision, early deve”
Convolutional Networks with Dense Connectivity (2019, background): “In [26], Deeply-Fused Nets (DFNs) were proposed to improve information flow by combining intermediate feature representations of different base networks.”
Deep Multi-Modal Object Detection and Semantic Segmentation for Autonomous Driving: Datasets, Methods, and Challenges (2019, background): “Alternatively, they can be fused hierarchically, such as by deep fusion [97], [161]:”
INet: Convolutional Networks for Biomedical Image Segmentation (2021, background): “Deeply fused nets [23] fuse intermediate representations of shallow layers as the input to deeper layers.”
AOGNets: Compositional Grammatical Architectures for Deep Learning (2017, methodology): “The Fractal Net [34] and deeply fused networks [60] provided an alternative way of implementing short path for training ultra-deep networks without residuals.”
AOGNets: Deep AND-OR Grammar Networks for Visual Recognition (2017, background): “enjoy repeated bottom-up/top-down feature exploration. Our AOGNet is created by intuitively simple yet principled grammars. It shares some spirit with the inception module [44], the deeply fused nets [49] and the DLA [54]. Grammar-like structures. A general framework of image grammar was proposed in”
Convolutional Neural Networks with Alternately Updated Clique (2018, methodology): “Multi-column [6] nets and Deeply-Fused Nets [38] also use fusion strategy and have a wide network structure.”
Training Shallow and Thin Networks for Acceleration via Knowledge Distillation with Conditional Adversarial Networks (2017, background): “Another line of research focuses on transferring intermediate features instead of soft targets from teacher to student [30, 41, 47, 44, 15, 49, 45].”
Densely Connected Convolutional Networks (2016, background): “In [38], Deeply-Fused Nets (DFNs) were proposed to improve information flow by combining intermediate layers of different base networks.”
Training Student Networks for Acceleration with Conditional Adversarial Networks (2018, background): “Another line of research focuses on transferring intermediate features instead of soft targets from teacher to student [14, 27, 36, 39, 40, 42].”
SORT: Second-Order Response Transform for Visual Recognition (2017, unknown): “This can be done by linearly summing them up [16], concatenating them [50], deeply fusing them [52], or using a bilinear [35] or recurrent [49] transform.”
Interleaved Group Convolutions for Deep Neural Networks (2017, methodology): “Summation [40, 38], average [22], and convolution operations [34, 4] following concatenation are often adopted to blend the outputs.”