WeightNet: Revisiting the Design Space of Weight Networks

By Ningning Ma, X. Zhang, Jiawei Huang. 2020

124 citations recorded by the source corpus.

We present a conceptually simple, flexible and effective framework for weight generating networks. Our approach is general that unifies two current distinct and extremely effective SENet and CondConv into the same framework on weight space. The method, called WeightNet, generalizes the two methods by simply adding one more grouped fully-connected layer to the attention activation layer. We use the WeightNet, composed entirely of (grouped) fully-connected layers, to directly output the convolutional weight. WeightNet is easy and memory-conserving to train, on the kernel space instead of the feature space. Because of the flexibility, our method outperforms existing approaches on both ImageNet and COCO detection tasks, achieving better Accuracy-FLOPs and Accuracy-Parameter trade-offs. The framework on the flexible weight space has the potential to further improve the performance. Code is available at this https URL.

Subfields: Computer Vision and Pattern Recognition, Model Architectures

How researchers see this paper

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

In Computer Vision and Pattern Recognition

1. Luc Van Gool
2. Lu Yuan
3. Dacheng Tao
4. Jianfeng Gao
5. Ming-Hsuan Yang
6. Xiaogang Wang
7. Zicheng Liu
8. Alan L. Yuille

In Model Architectures

1. Lu Yuan
2. Xiyang Dai
3. Dongdong Chen
4. Zicheng Liu
5. Shiji Song
6. Yinpeng Chen
7. Mengchen Liu
8. Gao Huang

Citations in the corpus

• Dynamic Neural Networks: A Survey (2021, background): “WeightNet [119] unifies the dynamic schemes of [13] and [20] by predicting the convolutional weights via simple grouped FC layers, achieving competitive results in terms of the accuracy-FLOPs4 and accuracy-parameters trade-offs.”
• Revisiting Dynamic Convolution via Matrix Decomposition (2021, methodology): “Its ability to achieve significant performance gains with negligible computational cost has motivated its adoption for multiple vision tasks (Su et al., 2020; Chen et al., 2020b; Ma et al., 2020; Tian et al., 2020).”
• Mobile-Former: Bridging MobileNet and Transformer (2021, methodology): “Main Results: Table 2 shows the comparison between Mobile-Former and classic efficient CNNs: (a) MobileNetV3 [14], (b) EfficientNet [26], and (c) ShuffleNetV2 [23] and its extension WeightNet [22].”
• Omni-Dimensional Dynamic Convolution (2022, methodology): “, 2020) or for convolutional weight modification (Ma et al., 2020; Lin et al., 2020; Quader et al., 2020).”