Adding attention blocks between ResNet-20 feature extractors gives small accuracy improvements on CIFAR-10 and MNIST, but the gains are within a range that needs error bars to trust.
Do vision trans- formers see like convolutional neural networks?, 2022
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
citation-role summary
background 1
citation-polarity summary
fields
cs.CV 1years
2024 1verdicts
CONDITIONAL 1roles
background 1polarities
support 1representative citing papers
citing papers explorer
-
CNNtention: Can CNNs do better with Attention?
Adding attention blocks between ResNet-20 feature extractors gives small accuracy improvements on CIFAR-10 and MNIST, but the gains are within a range that needs error bars to trust.