pre-trained models issue

[mostaan66]

Hi Caio,

I am trying to test your pre-trained "VGG16" model for the "active wave segmentation" using "predict.py" on my own data, but I'm facing some issues:

The output predicted label using "pred = M.predict(np.expand_dims(img/255, axis=0)) is a single value with shape (1, 1) while I assume it should return an array with the same size of the input image (i.e. assigning a single value prediction for each "pixel" and not the whole image). It seems to me that the pre-trained model is a classification model and not segmentation. In the end, this causes an error in proceeding running the codes and getting the predicted wave segmentation mask.

I also found that the input image to your pre-trained "VGG16" model is not 256x256 (as mentioned in the codes), instead I resized the input image to 128x128 and removed "img = gray2rgb(imread(image)) for my input image RGB images.

Could you please clarify it for me and if there is a mistake with your model, provide me with the correct pre-trained models for "wave segmentation"?

Many thanks!