JJGO / hyperlight

Modular and intuitive Hypernetworks in Pytorch
Apache License 2.0
32 stars 2 forks source link

Any example using detection, segmentation #2

Closed Ashutosh1995 closed 1 year ago

Ashutosh1995 commented 1 year ago

Hi,

Thanks for open-sourcing this repository!

My doubt is how to use the current repo if I want to train a normal segmentation network using hypernetwork. It would be great if you could provide some pointers.

Also, when running the following command,

hypernet = hl.Hypernet( input_shapes=hyperparam_shape, output_shapes=parameter_shapes, hidden_sizes=[16,64,128], )

I get the following error: Screenshot from 2023-05-18 22-28-53

Kindly help in this regard!

JJGO commented 1 year ago

Hey

There was a typo in the docs (Hypernet -> HyperNet). I've updated them and it should work now. You can also read them as an interactive colab here.

For a segmentation network, I'd suggest defining a regular U-Net model using pytorch and then using hyperlight to hypernetize the layers you want. The tutorial should be a good intro on how to do this.

Ashutosh1995 commented 1 year ago

Thanks @JJGO for the reply! Sure, I will try to implement it from my end.