| augment |
True, False |
block type="other" |
| block type |
"resnet", "other", "xception", "efficient" |
- |
| contrast factor |
"0.1", "0.0", "0.0", "0.1" |
augment=True |
| dropout |
"0.25", "0.0", "0.5" |
- |
| efficientnet version |
"b0", "b1", "b2", "b3", "b4", "b5", "b6", "b7" |
block type="efficient" |
| filters |
"16", "32", "64", "128", "256", "512" |
block type="other" |
| horizontal flip |
True, False |
augment=True |
| imagenet size |
True, False |
- |
| kernel size |
"3", "7", "5" |
block type="other" |
| learning rate |
"0.1", "0.01", "0.001", "0.0001", "1e-5", "2e-5" |
- |
| max pooling |
True, False |
block type="other" |
| normalize |
True, False |
- |
| num blocks |
"1", "2", "3" |
block type="other" |
| num layers |
"1", "2" |
block type="other" |
| optimizer |
"sgd", "adam", "adam weight decay", "ndam" |
- |
| efficientnet pretrained |
False, True |
block type="efficient" |
| resnet pretrained |
False, True |
block type="resnet" |
| xception pretrained |
False, True |
block type="xception" |
| reduction type |
"flatten", "global avg", "global max" |
block type="resnet"/"xception"/"efficient" |
| resnet version |
"resnet50", "resnet152", "resnet101", "resnet50 v2", "resnet101 v2", "resnet152 v2" |
block type="resnet" |
| rotation factor |
"0.1", "0.0", "0.0", "0.1" |
augment=True |
| separable |
True, False |
- |
| efficientnet trainable |
True, False |
efficientnet pretrained=True |
| resnet trainable |
True, False |
resnet pretrained=True |
| xception trainable |
True, False |
xception pretrained=True |
| translation factor |
"0.0", "0.1", "0.1", "0.0" |
augment=True |
| vertical flip |
True, False |
augment=True |
| zoom factor |
"0.0", "0.1", "0.1", "0.0" |
augment=True |
| double step |
True, False |
trainable=True |
| triple step |
True, False |
Multi step=True |
| epoch ratio |
"0.1", "0.2", "0.3", "0.4", "0.5" |
Multi step=True |
| lr ratio |
"1.0", "0.1", "0.01" |
Multi step=True |
| freeze |
"all", "no", "bn" |
Multi step=True |
| momentum |
"0.0", "0.1", "0.5", "0.9", "0.99" |
optimizer="sgd" |
| end learning rate |
"1e-4", "1e-5", "2e-6", "1e-6", "0.0" |
optimizer="adam weight decay" |
| weight decay rate |
"0.001", "0.005", "0.01", "0.05", "0.1" |
optimizer="adam weight decay" |
| initializer |
"he uniform", "lecun uniform","glorot uniform" |
block type="other" |
| activation |
"selu", "tanh","relu" |
block type="other" |