vdumoulin/conv_arithmetic ? reverse-engineered prompt

I want to create a visual guide that explains convolution arithmetic in deep learning. The main goal is to generate a bunch of clear, animated GIFs that make the concepts easy to grasp.

You should create animations for standard convolutions, transposed convolutions, and dilated convolutions. For each type, I want to see examples of how different parameters change the output, so show variations with different padding (none, some, full) and strides. In the animations, please use different colors to clearly distinguish between the input and output maps, just like in the examples you can find online for this topic.

Finally, take all these generated animations and embed them in a professional-looking technical report. The final output should be a compiled PDF, properly formatted like an academic paper with a bibliography. Use TeX to lay out the document.