Transfer Learning Adapter Skill for Frozen Vision Backbones
Freeze a pretrained backbone, train a new head on owned synthetic images, and verify which parameters changed.
中文Install in Claude Code
Use Python, PyTorch, a permitted pretrained model, and synthetic circle and square packaging labels.
- Get the complete `plugins/ai-ml/transfer-learning-adapter/skills/adapting-transfer-learning-models` directory from the pinned GitHub commit.
- Copy it into the target project's `.claude/skills/`. Check `.claude/skills/adapting-transfer-learning-models/SKILL.md`, start Claude Code at the project root, and call `/adapting-transfer-learning-models`.
Example task
Keep the backbone fixed, train the two-class head, compare before-and-after summaries, and evaluate held-out images.What to check
The backbone parameters remain the same and the head changes. The evaluation lists per-class support and results.
Original SKILL.md on GitHub · Updated
Frequently asked question
Does freezing parameters also freeze BatchNorm buffers?
Check module mode. Keep the frozen backbone in the appropriate evaluation mode and verify its buffers when they must stay fixed.