search

deephealthproject / use-case-pipelines

Use case pipelines based on EDDL and ECVL libraries. Different tasks (e.g. classification and segmentation) and datasets (e.g. MNIST, ISIC, and PNEUMOTHORAX) are taken into account.
Other
2 stars 5 forks source link

ResNet50 is missing in the settings options #18

Closed abadouh closed 2 years ago

abadouh commented 3 years ago

In order to run UC12 classification training application, the code call to setting with ResNet50.

https://github.com/deephealthproject/use-case-pipelines/blob/57f9c9deaedd4048627d19004b6e6dd5bc069858/cpp/skin_lesion_classification_training.cpp#L23

As we can see below, ResNet50 is not part of the models.

https://github.com/deephealthproject/use-case-pipelines/blob/57f9c9deaedd4048627d19004b6e6dd5bc069858/cpp/utils/utils.cpp#L89

I made the changes in the source code and added the ResNet50 option.

The application now failing to build the model: terminate called after throwing an instance of 'AsymmetricPaddingException' what(): In layer conv2d1: Padding asymmetry detected (top=2, bottom=3, left=2, right=3). The padding asymmetry is not allowed in a Conv layer, we suggest you to use an explicit padding layer before this layer to fix the asymmetry.

abadouh commented 2 years ago

In the attachment you can find fix for the padding exception.

Tested with ECVL0.4.0+EDDL1.0.2a

resnet_model.txt

lauracanalini commented 2 years ago

Fixed in a4e6f09febd8caf6e293d1c395ec6205fa5bc355