Agent skill · NVIDIA
tao-train-visual-changenet
Visual ChangeNet for binary image classification and segmentation in AOI defect detection. Use when training,
What it needs
About 9k tokens when loaded.
What this skill does
Visual ChangeNet Standalone install? If this session was not initialized by the TAO skill bank plugin, run the tao-setup skill first (host preflight, credentials, cross-skill discovery). Visual ChangeNet is a TAO Toolkit model for visual inspection and defect detection. It supports two tasks: Classify — Binary image classification using a siamese-style architecture with a shared backbone (C-RADIO ViT) and a learnable difference module. Compares image pairs to classify defects as PASS/NOPASS. Segment — Pixel-level change segmentation using a ViT-Large NVDINOv2 backbone. Compares before/after image pairs to produce a binary change mask. The backbone weight (cradiov2vitbasepatch16224) is the public nvidia/C-RADIOv2-B model from HuggingFace, distributed as model.safetensors (~393 MB). The TAO container does not auto-fetch from HF URLs — ptmutils.loadpretrainedweights() hands the pretrainedbackbonepath value to torch.load(path) / safetensors.torch.loadfile(path) directly. Passing an https://huggingface.co/... URL or a repo id produces FileNotFoundError and the run fails with Execution status: FAIL within a few seconds. Stage the file locally before launch with the bundled helper (idempotent — reuses an already-staged file): This is a public download — no NGC CLI, no NGC org, and no credentials are required, and there is no ngc:// transfer-learning checkpoint dependency (VCN trains from this backbone). Run it in the CPU shell, where host network and HFTOKEN live; HFTOKEN is read only if set (gated mirror / rate limit) and is needed at staging time only, never inside the training container. Mount the staged file into the container (-v <workspace>/pretrainedmodels/C-RADIOv2B.safetensors:/data/pretrainedmodels/C-RADIOv2B.safetensors) and set the spec model.backbone.pretrainedbackbonepath to the container path. Segment specs use model.backbone.type: vitlargenvdinov2 and the NVDINOv2 checkpoint family. …
How to use it
Reference it in AdaL, Claude Code, Cursor or any coding agent — nothing to install:
@skills NVIDIA/tao-train-visual-changenet