How Much Position Information Do Convolutional Neural Networks Encode?
arXiv:2001.08248
Abstract
In contrast to fully connected networks, Convolutional Neural Networks (CNNs) achieve efficiency by learning weights associated with local filters with a finite spatial extent. An implication of this is that a filter may know what it is looking at, but not where it is positioned in the image. Information concerning absolute position is inherently useful, and it is reasonable to assume that deep CNNs may implicitly learn to encode this information if there is a means to do so. In this paper, we test this hypothesis revealing the surprising degree of absolute position information that is encoded in commonly used neural networks. A comprehensive set of experiments show the validity of this hypothesis and shed light on how and where this information is represented while offering clues to where positional information is derived from in deep CNNs.
Accepted to ICLR 2020
References in corpus (3)
Cited by in corpus (28)
- PVT v2: Improved Baselines with Pyramid Vision Transformer
- Alias-Free Generative Adversarial Networks
- SOLOv2: Dynamic and Fast Instance Segmentation
- Conditional Positional Encodings for Vision Transformers
- ResT: An Efficient Transformer for Visual Recognition
- Counting Varying Density Crowds Through Density Guided Adaptive Selection CNN and Transformer Estimation
- ObjFormer: Learning Land-Cover Changes From Paired OSM Data and Optical High-Resolution Imagery via Object-Guided Transformer
- Position, Padding and Predictions: A Deeper Look at Position Information in CNNs
- A Survey of Visual Transformers
- SOLO: Segmenting Objects by Locations
- Mind the Pad -- CNNs can Develop Blind Spots
- PLADE-Net: Towards Pixel-Level Accuracy for Self-Supervised Single-View Depth Estimation with Neural Positional Encoding and Distilled Matting Loss
- Diffusion-Based Audio Inpainting
- Positional Encoding as Spatial Inductive Bias in GANs
- NeRD: Neural Representation of Distribution for Medical Image Segmentation
- Shift Equivariance in Object Detection
- PiSLTRc: Position-informed Sign Language Transformer with Content-aware Convolution
- A Comprehensive Study of Data Augmentation Strategies for Prostate Cancer Detection in Diffusion-weighted MRI using Convolutional Neural Networks
- Breaking Shortcut: Exploring Fully Convolutional Cycle-Consistency for Video Correspondence Learning
- Multi-Scale Local-Temporal Similarity Fusion for Continuous Sign Language Recognition
- SOLO: A Simple Framework for Instance Segmentation
- GANSeg: Learning to Segment by Unsupervised Hierarchical Image Generation
- Tool as Embodiment for Recursive Manipulation
- Neural Mesh Refiner for 6-DoF Pose Estimation
- Do CNNs Encode Data Augmentations?
- An Empirical Method to Quantify the Peripheral Performance Degradation in Deep Networks
- Ripple Attention for Visual Perception with Sub-quadratic Complexity
- Isometric Graph Neural Networks