Pix2Pix model examples¶
Experimental interface — under construction
XWhy currently exports Pix2PixExplainer, but its explain() method raises NotImplementedError. The examples described here are a documentation roadmap, not an executable workflow.
Pix2Pix-style models are one example of conditional image-to-image generation. They belong under Image Generation → Image Editing because the model transforms a source image into a target image rather than returning a classification score.
Planned first example¶
A future worked example should document:
- the source image and target transformation task;
- the Pix2Pix model, weights, and preprocessing;
- the source-image regions selected for perturbation;
- the output-distance or perceptual-similarity measure;
- the local surrogate configuration;
- the generated attribution map;
- fidelity, stability, and runtime evidence;
- limitations of interpreting generative outputs.
Planned comparison cases¶
- removing or masking source-image regions;
- perturbing the conditioning input;
- comparing pixel-space and perceptual distances;
- checking attribution stability across generation seeds;
- detecting unintended changes outside the target edit region.