A novel feature-scrambling approach reveals the capacity of convolutional neural networks to learn spatial relations.
Amr FarahatFelix EffenbergerMartin A. VinckPublished in: CoRR (2022)
Keyphrases
- spatial relations
- convolutional neural networks
- spatial reasoning
- spatial information
- spatial relationships
- segmented regions
- spatial configuration
- closely related
- temporal relations
- spatial objects
- image features
- convolutional network
- spatial knowledge
- physical objects
- object parts
- text mining
- special case
- pairwise
- directional relations
- feature extraction