titu1994/keras-squeeze-excite-network
Keras implementation of Squeeze-and-Excitation blocks and SE-enhanced versions of ResNet, Inception, DenseNet, and other popular CNN architectures.
Not currently ranked — collecting fresh signals.
star history
This repository provides Keras 2.0+ implementations of Squeeze-and-Excitation (SE) blocks, a channel attention mechanism for convolutional neural networks introduced in the 2017 SE-Net paper. It includes prebuilt SE-enhanced model architectures including SEResNet, SE-InceptionV3, SE-Inception-ResNet-v2, SE-ResNeXt, SE-MobileNets, and SE-DenseNet variants. The SE block applies global average pooling, dense layers, and sigmoid activation to learn channel-wise attention weights that recalibrate feature map responses.
Frequently asked
- What is titu1994/keras-squeeze-excite-network?
- Keras implementation of Squeeze-and-Excitation blocks and SE-enhanced versions of ResNet, Inception, DenseNet, and other popular CNN architectures.
- Is keras-squeeze-excite-network open source?
- Yes — titu1994/keras-squeeze-excite-network is open source, released under the MIT license.
- What language is keras-squeeze-excite-network written in?
- titu1994/keras-squeeze-excite-network is primarily written in Python.
- How popular is keras-squeeze-excite-network?
- titu1994/keras-squeeze-excite-network has 401 stars on GitHub.
- Where can I find keras-squeeze-excite-network?
- titu1994/keras-squeeze-excite-network is on GitHub at https://github.com/titu1994/keras-squeeze-excite-network.