adapt pre-trained models to a Self-ONN converted network

[SarinaAhmadi]

When I use Self-ONN(q>1) instead of Conv in the Alexnet network, I get this error:

How I should load and adapt pre-trained models to a Self-ONN converted network?

[junaidmalik09]

It is not currently possible to transfer convolutional weights to a Self-ONN network. When adapting pre-trained models, one strategy could be to keep the earlier convolutional layers while replacing the layers towards the end of the network with Self-ONNs. The Self-ONN layers can then be trained from scratch while the convolutional layers can be frozen or fine-tuned, depending on the need. Please note that this is still an open topic and requires further study.