The official PyTorch implementation of "DGCA: A dual global context attention module for image recognition".
Figure 1: An overview of the proposed DGCA module.
Description. Structurally, the DGCA module consists of two parallel branches, in which the upper branch focuses on capturing global contextual information in the spatial dimension, while the lower branch is responsible for modeling long-range semantic dependencies in the channel dimension.
DGCA. Detail of implementations, including modules and the networks, can be found in Cifar and ImageNet in this repository.
Figure 2: Details of the proposed GSA module.
Figure 3: Detailed illustration of the proposed GCA module.
- OS: Ubuntu 18.04.1
- CUDA: 11.6
- Python: 3.9.12
- Toolkit: PyTorch 1.10
- GPU: RTX A6000 (4x)
- ptflops
- For generating GradCAM++ results, please follow the code on this repository
Figure 4: Illustration of the DGCA module integrated into CNNs. Here, (a) CNNs without skip connections; (b) DGCA module integrated into (a); (c) CNNs with skip connections; (d) DGCA module integrated into (c).
Table 1: Comparison of classification performance of various attention methods on CIFAR-100 based on ResNet-50 architecture.
Figure 5: Comparison of our method with other state-of-the-art attention methods.
Table 2: Comparison of efficiency (i.e., Parameters and FLOPs) and effectiveness (i.e., Top-1/Top-5 Acc) of different attention methods on ImageNet-1K classification using ResNet with 18, 34, 50, and 101 layers as backbones, respectively.
Figure 6: Comparison of classification performance on ImageNet-1K between our method and other attention methods.
If you find DGCA useful in your research, please consider citing:
If you have any suggestion or question, you can leave a message here or contact us directly: yang_y9802@163.com. Thanks for your attention!