Delving Deeper into Anti-Aliasing in ConvNets

By Xueyan Zou, Fanyi Xiao, Zhiding Yu, Yong Jae Lee. 2020

148 citations recorded by the source corpus.

Aliasing refers to the phenomenon that high frequency signals degenerate into completely different ones after sampling. It arises as a problem in the context of deep learning as downsampling layers are widely adopted in deep architectures to reduce parameters and computation. The standard solution is to apply a low-pass filter (e.g., Gaussian blur) before downsampling (Zhang in: ICML, 2020). However, it can be suboptimal to apply the same filter across the entire content, as the frequency of feature maps can vary across both spatial locations and feature channels. To tackle this, we propose an adaptive content-aware low-pass filtering layer, which predicts separate filter weights for each spatial location and channel group of the input feature maps. We investigate the effectiveness and generalization of the proposed method across multiple tasks, including image classification, semantic segmentation, instance segmentation, video instance segmentation, and image-to-image translation. Both qualitative and quantitative results demonstrate that our approach effectively adapts to the different feature frequencies to avoid aliasing while preserving useful information for recognition. Code is available at https://maureenzou.github.io/ddac/ .

Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition

How researchers see this paper

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

In Artificial Intelligence

In Computer Vision and Pattern Recognition

Citations in the corpus

On Buggy Resizing Libraries and Surprising Subtleties in FID Calculation (2021, background): “While it was replaced by operations such as max-pooling, based on empirical performance [65], recent works have demonstrated that antialiasing can be compatible and improve performance in convolutional networks [79,85], transformers [57], NeRFs [4], and GANs [36].”
Global Pooling, More than Meets the Eye: Position Information is Encoded Channel-Wise in CNNs (2021, neutral): “Zou et al. [35] extended this work by using content-aware learned ﬁlter weights, which predict separate ﬁlter weights for each spatial and channel location in the input.”
Alias-Free Generative Adversarial Networks (2021, background): “Several recent works have studied the lack of translation equivariance in CNNs, mainly in the context of classification [27, 34, 63, 5, 37, 57, 65, 12, 67].”
Blending Anti-Aliasing into Vision Transformer (2021, methodology): “The model beneﬁts from an early anti-aliasing operation right after the self-attention computation, which is distinctive from the observations in CNNs [10, 12].”
Uncertainty-Aware Learning for Zero-Shot Semantic Segmentation (2020, background): “As a fundamental problem in computer vision, this task has attracted a lot of attention from the research community and achieved great success along with the development of deep learning in recent years [7, 8, 14, 15, 19, 24, 38, 45, 50, 51, 57, 59, 60, 63, 66].”
On Aliased Resizing and Surprising Subtleties in GAN Evaluation (2022, background): “While it was replaced by operations such as max-pooling, based on empirical performance [60], recent works have demonstrated that antialiasing can be compatible and improve performance in convolutional networks [73,79], transformers [52], NeRFs [3], and GANs [32].”