Tensorfllow: load checkpoint from changed model

Question

For some reason I want to test the difference in the performance of a detector and his identical version but finetuned with some 3d convolutions.
The model of the detector is google EfficientDet, the weights are finetuned on custom data.
I was wondering if it was possible to load my custom weight in a model in which the graph-def is not the same (there would be 3d convolutions at some layers). And what could be the way to do that.
I'm new to Tensorflow and a bit sad because in Pytorch this would be so easy

Thanks

Andrey · Accepted Answer · 2021-01-17 06:02:31Z

1

You can load the second model, get weights of the layer and set weights of your model:

source_model = keras.models.load_model('path/to/location')
weight = source_model.layers[0].get_weights()   # <= change index here
EfficientDetModel.layers[0].set_weights(weight) # <= change index here

answered Jan 17, 2021 at 6:02

Andrey

6,6393 gold badges24 silver badges51 bronze badges

Sign up to request clarification or add additional context in comments.

Collectives™ on Stack Overflow

Tensorfllow: load checkpoint from changed model

1 Answer 1

Comments

Your Answer

Hot Network Questions

Collectives™ on Stack Overflow

1 Answer 1

Comments

Your Answer

Sign up or log in

Post as a guest

Related