I have 2 different models with each model doing a separate function and have been trained with different weights. Is there any way I can merge these two models to get a single model.
If it can be merged
- How should I go about it? Will the number of layers remain the same?
- Will it give me any performance gain?(Intuitively speaking, I should get a higher performance)
- Will the hardware requirements change when using the new model?
- Will I need to retrain the model? Can I somehow merge the trained weights?
If the models cannot be merged
- Why so? After all, convolution is finding the correct pattern in data.
- Also, if CNN's cannot be merged, then how do skip-connections like ResNet50 work?
What I currently have
Image ---(model A) ---> Temporary image ---(Model B)---> Output image
What I want:
Image ---(model C) ---> Output image