I have a .pb file alone created using freeze_graph.py. Other than the standalone .pb file i don’t have any files in the folder. Till now i have figured out a way to visualize the model using tensor board and to fetch the tensors involved in the Net. Also the .pb was a model created using ResNet architecture.
1.How do i make use of this file to make predictions on the custom data which i have? If i could have a process pipeline it would be off great help.
2.Also how do i use the .pb file in keras?
Advertisement
Answer
Well you could just load in the frozen model and then define an input pipeline. Here’s an article that describes this in more detail. I would also suggest to have a look at Tensorflow Serving.
Don’t think that this makes sense as Keras uses HDF5 to store their model. Also it doesn’t make any sense to use the single .pb for development as this is not its use case. If you want to have flexibility, you should use the checkpoints which has the graph and the weights for certain epochs.