We add attention mechanism come from Squeeze-and-Excitation and Selective Kernel to InvertedResidual block in shuffleNetV2.
which can be found in the dir about 'shuffle_SE' and 'shuffle_SK'
the main directory build as
proto_1_vggnet
proto_2_googlenet
proto_3_resnet
proto_4_densenet
proto_5_mobilenet
remix_1_shufflenet
shuffle_SE
shuffle_SK
model.py
predict.py
train.py
The difference between these Net comes in 'model.py'
Additionally, we backup the train info in the '/draw_plot', Hope these a bit help you understand the code.
Cervical cancer screening dataset used for classification located in '\data\cancer_data', which split into train and val.
We fetch 10 example cervical cancer image in each categories(Although these pictures not enough to support training).
If you need a complete cervical cancer dataset or make a deeper academic exchanges, please send an email to [email protected].