Layered view#

layered_view represents layers as scaled boxes and is especially useful for convolutional architectures. Controls include minimum/maximum rendered sizes, x/y/z scaling, layer type/index ignore lists, color maps, spacing, padding, legends/text, and output file handling.

Visual dimensions are explanatory encodings, not literal physical tensor sizes. Extreme tensor dimensions are intentionally clipped/scaled to keep the figure usable. When exact shapes matter, include text labels or a separate model summary.