Mayukh Deb

What exactly is -use_fft doing ?

2

Hi @ProGamerGov, I've been digging around quite a lot lately to understand the significance of `-use_fft`. I'm trying to implement it myself on my own feature visualization library. What I...

add support for noisegrad

2

The paper can be found here: https://arxiv.org/pdf/2106.10185.pdf The core idea is to add noise to the weights on each iteration, in proportion to the weights variance in each layer. It...

instead of using: ```python layers_to_use = [model.layer1, model.layer2] ``` and then accessing them in the custom function as: ```python layer_outputs[0] and layer_outputs[1] ``` we can define `layers_to_use` as a dict...

When calling `dreamer.caricature()`, it does not consider the fact that `self.__custom_normalization__` is not `None` and hence doesnt even use it before the forward pass.

all the layer outputs can be stored in a dict instead of a list. it could look like: ```python layer_outputs = { '__input_image__': input_image_tensor, 'conv3.branch1': some_tensor, 'fc': another_tensor } ````...

Not sure but it happens when we call: ```python dreamy_boi.set_custom_transforms(transforms = my_transforms) ``` before ```python dreamy_boi.set_custom_normalization( normalization_transform = my_normalization_transform() ) ```

something like: ```python from torch_dreams.n_channel_param import NChannelParam param = NChannelParam( num_channels = 2, mean = [0.5,0.5], std = [0.5, 0.5], device = 'cuda' ) out = dreamy_boi.render( layers = [model.layer3],...

This might contain stuff which generally require a large amount of boilerplate in notebooks/scripts, but are used often in code Something like: ```python from torch_dreams.recipes.losses import SnapshotLoss ``` Obviously inspired...

This might be done as: ```python torch.save(image_parameter.state_dict(), 'param.pt') image_param.load_state_dict(torch.load('param.pt')) ```