REVIEW 6 cited by
Making Convolutional Networks Shift-Invariant Again
Not yet reviewed by Pith; the record is open.
This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.
SPECIMEN: schema-true, not a live event
T0 review · schema-true
One-sentence machine reading of the paper's core claim.
pith:XXXXXXXX · record.json · timestamp
Signed reviews
read the original abstract
Modern convolutional networks are not shift-invariant, as small input shifts or translations can cause drastic changes in the output. Commonly used downsampling methods, such as max-pooling, strided-convolution, and average-pooling, ignore the sampling theorem. The well-known signal processing fix is anti-aliasing by low-pass filtering before downsampling. However, simply inserting this module into deep networks degrades performance; as a result, it is seldomly used today. We show that when integrated correctly, it is compatible with existing architectural components, such as max-pooling and strided-convolution. We observe \textit{increased accuracy} in ImageNet classification, across several commonly-used architectures, such as ResNet, DenseNet, and MobileNet, indicating effective regularization. Furthermore, we observe \textit{better generalization}, in terms of stability and robustness to input corruptions. Our results demonstrate that this classical signal processing technique has been undeservingly overlooked in modern deep networks. Code and anti-aliased versions of popular networks are available at https://richzhang.github.io/antialiased-cnns/ .
Forward citations
Cited by 6 Pith papers
-
From Spatial to Spectral: An Efficient, Frequency-Guided Feature Representation Learner for Small Object Detection
Proposes DERNet with Decompose-Enhance-Reconstruct operator and three plug-and-play modules to shift small object detection from spatial to spectral feature processing, claiming better performance than YOLOv11 with 1/...
-
Learning to Adapt to Position Bias in Vision Transformer Classifiers
A vision transformer can learn a single scalar that gates its position embedding, and a new SHAP-based metric measures when that gate should be small or large.
-
Geometrically Constrained and Token-Based Probabilistic Spatial Transformers
A probabilistic, component-wise spatial transformer using a shared frozen tokenizer and augmentation-guided alignment loss improves geometric robustness in fine-grained moth classification.
-
A Spatial-Frequency Aware Multi-Scale Fusion Network for Real-Time Deepfake Detection
SFMFNet uses wavelet-frequency gating, token-selective cross-attention, and blur pooling to reach 0.8682 average cross-dataset AUC with only 1.27 GFLOPs and 6.64M parameters.
-
Faithful, Interpretable Chest X-ray Diagnosis with Anti-Aliased B-cos Networks
Combining B-cos networks with anti-aliasing pooling (FLC or BlurPool) reduces grid artifacts in chest X-ray explanation maps while keeping diagnostic accuracy close to baseline networks.
-
From Machine Learning to Large-Scale EO Products: Best Practices for Making Maps
This paper is a concise best-practices guide for producing large-scale machine learning maps from Earth observation data, spanning preprocessing, training, uncertainty, production, and validation.
Discussion (0). Continue with ORCID to comment.