DeepDream | Nice little modular interface for running deep dreams | Computer Vision library

So, I wrote the code below many months ago, and it's worked pretty well. Though I am struggling on how I can simplify it and make it more efficient.

The functions below split an image tensor (B, C, H, W) into equal sized tiles (B, C, H, W) and then you can do stuff individually to the tiles in order to save memory. Then when rebuilding the tensor from the tiles, it uses masks to ensure that the tiles are seamlessly blended back together. The 'special masks' in the masking function handle when tiles in the right most column or tiles in the bottom row can't use the same overlap as the other tiles. This means that the right edge tiles and the bottom tiles may sometimes have almost none of their content visible. This is done to ensure that the tiles are always the exact specified size, regardless of the original image/tensor's size (important for visualization/DeepDream, neural style transfer, etc...). The adjacent row/column to the edge row/column also has special masks as well for where they overlap with the edge row/column.

There are 8 possible masks for every tile, and 4 of those masks can be used at once. The 4 possible masks are left, right, top, and bottom, with a special version for each mask.

I'd like to randomly rotate an image tensor (B, C, H, W) around it's center (2d rotation I think?). I would like to avoid using NumPy and Kornia, so that I basically only need to import from the torch module. I'm also not using torchvision.transforms, because I need it to be autograd compatible. Essentially I'm trying to create an autograd compatible version of torchvision.transforms.RandomRotation() for visualization techniques like DeepDream (so I need to avoid artifacts as much as possible).

I need to calculate the covariance matrix for RGB values across an image dataset, and then apply Cholesky decomposition to the final result.

The covariance matrix for RGB values is a 3x3 matrix M, where M_(i, i) is the variance of channel i and M_(i, j) is the covariance between channels i and j.

The end result should be something like this:

The curl API request that I am trying to transform into Swift:

I'm following the tutorial here:

in order to create a python program that will create a deep-dream style img and save in onto disk. I thought that changes to the following lines should do the trick:

I followed this tutorial and used the ipynb notebook from the Github page to generate deepdream images in Google Colaboratory. The tutorial uses the Inception5h network. 12 layers in this model are commonly used for generating images.

Each layer consists of approximately 500 individual features, which recognize different patterns. It is possible to select specific features in a layer, which yield different results. I've generated images of each feature in layer 6, 'mixed4a:0'. What I'm trying to do now is mix these features.

A specific layer is selected like this:

So far I have been able to ''manually'' process images by replacing 'picture' in

In the deep dream example using tensorflow here, the code references the inception5h model developed by google. However the original code from google here is using caffe, not tensorflow, probably because tensor flow did not exist then. How is it that the same model can be used by two different frameworks? The 'deploy.prototxt' distributed with the bvlc_googlenet.caffemodel lists many convolution layers but the tensor flow implementation of the same model does not reference them and seems to use many fewer layers.

If I get a pretained model without a 'deploy.prototxt' file, how can i determine how many layers the model has and how to reference them?

I am looping through a series of names in a tuple and I want to save the output during each loop using the tuple data as the filename. However the names have slashes in them.