---

name: spikevision description: "Legacy neuromorphic dataset wrappers and transforms for snnTorch; deprecated in favor of Tonic and kept for compatibility, import smoke, and local file handling only." metadata: disco-role: operating disable-model-invocation: true license: MIT

spikevision

[!WARNING] snntorch.spikevision is deprecated. For new neuromorphic dataset workflows, use Tonic. This sub-skill exists only for legacy compatibility, import/signature smoke, and local file handling.

Covers

• Legacy dataset wrappers: NMNIST, DVSGesture, SHD
• Legacy base helpers: NeuromorphicDataset, StandardTransform
• Local transform helpers and file readers used by the legacy datasets
• Lightweight import/signature inspection without dataset downloads

Does not cover

• Tonic-first workflows
• Training notebooks or model training recipes
• Plotting helpers
• NIR interoperability
• Real dataset downloads during extraction

Routing rule

• If the request is for a modern neuromorphic pipeline, route to Tonic guidance first.
• If the request is to support, inspect, or migrate an old spikevision script, use the bundled references and the introspection helper.
• If the task would require fetching raw datasets from the network, stop and treat that as out of scope for extraction.

Primary entry points

• Read references/api-reference.md for supported symbols, signatures, and on-disk expectations.
• Read references/workflows.md for safe legacy usage patterns.
• Read references/troubleshooting.md for dependency, layout, and constructor quirks.
• Run scripts/spikevision_introspect.py to confirm the deprecation warning, exposed classes, and safe transform constructors.

Working constraints

• Keep this surface legacy-only.
• Prefer wrapper-level compatibility over adding new behavior.
• Keep dataset downloads and cache creation out of synthetic verification.
• Use the modern Tonic route whenever a user is starting fresh.