Keras/Tensorflow Mask-rcnn custom dataset training


I got assigned a project for my thesis where i should code a programm to detect trees (or more specific treetrunks). So I went and took some pictures to create a dataset for deep-learning. I created the annotations file and used a Notebook by Pysource to create the model. The problem now seems that this Notebook only gives me the Keras weights (h5). When i tried to convert the file to Tensorflow (pb), to implement it into my Python program. But it seems the converter needs the config to convert it into TF. I already searched on here for some solutions but the only answer i see it to use and not model.save_weights(). But i can’t find this line of code in the notebook to change it. The function model.load_weights() for just loading the weights doesn’t work either because it says that the Model class doesn’t have that function.Is there maybe any other way to convert the keras into tf or any other notebook that gives me the weights+config? Or a way to directly use the Keras weights only in a Python program for instance segmentation without converting it to TF? This is the notebook I was using: Google Colab

Thanks in advance!