How to create a new network that has the same backbone but two detection heads: one with the imagenet weights and the second head with the new custom classes. I want to predict both imagenet and custom classes with the model

[vijay2712]