Summary of Attention Augmented Convolutional Networks, by Irwan Bello et al.
Attention Augmented Convolutional Networks
by Irwan Bello, Barret Zoph, Ashish Vaswani, Jonathon Shlens, Quoc V. Le
First submitted to arxiv on: 22 Apr 2019
Categories
- Main: Computer Vision and Pattern Recognition (cs.CV)
- Secondary: None
GrooveSquid.com Paper Summaries
GrooveSquid.com’s goal is to make artificial intelligence research accessible by summarizing AI papers in simpler terms. Each summary below covers the same AI paper, written at different levels of difficulty. The medium difficulty and low difficulty versions are original summaries written by GrooveSquid.com, while the high difficulty version is the paper’s original abstract. Feel free to learn from the version that suits you best!
| Summary difficulty | Written by | Summary |
|---|---|---|
| High | Paper authors | High Difficulty Summary Read the original abstract here |
| Medium | GrooveSquid.com (original content) | Medium Difficulty Summary Convolutional networks have been the standard choice for computer vision applications, but they have limitations when it comes to capturing global information. Self-attention has shown promise in sequence and generative modeling tasks, so researchers investigated using self-attention for discriminative visual tasks instead of convolutions. The paper proposes a novel 2D relative self-attention mechanism that can replace convolutions as a standalone primitive for image classification. Experiments show that combining both convolutions and self-attention yields the best results. To augment convolutional operators, researchers concatenated feature maps from convolutions with those from self-attention. This Attention Augmentation method leads to consistent improvements in image classification on ImageNet and object detection on COCO across various models and scales, including ResNets and a state-of-the-art mobile constrained network. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Computer vision is the ability of computers to recognize images. For a long time, researchers have used something called convolutional networks to do this. But these networks have some limitations. They can only look at small parts of an image, not the whole thing. Recently, another way called self-attention has been developed to help with this problem. In this paper, scientists are trying to use self-attention for computer vision tasks instead of convolutional networks. They came up with a new way to do this that works well for recognizing objects in images. It’s called Attention Augmentation. This method helps computers recognize objects better and faster than before. |
Keywords
* Artificial intelligence * Attention * Image classification * Object detection * Self attention
