---

name: methods-and-api description: "Routes expert pytorch-grad-cam method selection, public signatures, lifecycle, backend, and edge-case API troubleshooting." disable-model-invocation: true metadata: disco-role: operating license: MIT

Methods and API

Use this sub-skill when the user asks which CAM method to choose, what a class signature or default parameter is, why a method behaves differently, how the context manager and hook lifecycle work, or how advanced methods such as FinerCAM, SegEigenCAM, ShapleyCAM, KPCA-CAM, or AblationCAM variants differ.

Read first

• references/methods-reference.md for the method catalog and selection guidance.
• references/api-signatures.md for the verified installed constructor signatures.
• references/backend-and-lifecycle.md for context manager, hook, and device behavior.
• references/troubleshooting.md for expert API mistakes and backend-specific gotchas.

Common tasks

• Choose a method for speed, faithfulness, or transformer compatibility.
• Check an installed constructor signature before writing a prompt or helper.
• Understand why a method needs comparison_categories, alpha, ablation_layer, or a reshape_transform.
• Debug FinerCAM on a binary/ternary classifier or check SVD projection helpers for mutation-free behavior.

Distinguish from other sub-skills

• Use cam-generation for the ordinary end-to-end workflow of producing and overlaying heatmaps.
• Use model-task-adaptation for custom targets and reshape transforms.
• Use metrics-and-evaluation for ROAD, ARCC, RefineCAM, and DFF.