visual-attention-tf 1.2.0

CNN Attention layer to be used with tf or tf.keras

Visual_attention_tf

A set of image attention layers implemented as custom keras layers that can be imported dirctly into keras

Currently Implemented layers:

• Pixel Attention : Efficient Image Super-Resolution Using Pixel Attention(Hengyuan Zhao et al)
• Channel Attention : CBAM: Convolutional Block Attention Module(Sanghyun Woo et al)
• Efficient Channel Attention : ECA-Net: Efficient Channel Attention for Deep Convolutional Neural Networks

Installation

You can see the projects official pypi page : https://pypi.org/project/visual-attention-tf/

pip install visual-attention-tf

Use --no-dependencies if you have tensorflow-gpu installed already

Usage:

from tensorflow.keras.models import Model
from tensorflow.keras.layers import Input, Conv2D
from visual_attention import PixelAttention2D , ChannelAttention2D,EfficientChannelAttention2D

inp = Input(shape=(1920,1080,3))
cnn_layer = Conv2D(32,3,,activation='relu', padding='same')(inp)

# Using the .shape[-1] to simplify network modifications. Can directly input number of channels as well
Pixel_attention_cnn = PixelAttention2D(cnn_layer.shape[-1])(cnn_layer)
Channel_attention_cnn = ChannelAttention2D(cnn_layer.shape[-1])(cnn_layer)
EfficientChannelAttention_cnn = EfficientChannelAttention2D(cnn_layer.shape[-1])(cnn_layer)