Train models to spot defects in image pairs.
What it does
Trains and runs image-pair inspection models that classify defects or produce pixel-level change masks.
When to use it
Use for training, evaluation, export, or inference involving PCB defects, visual inspection, or change segmentation.
How to use it
Give it an inspection dataset and a training or inference request; it produces model artifacts, metrics, or predictions in result files.
What you provide
- Your files
Uses
Good to know
- a credential is hard-coded: encryption_key
What you need · 8
Requires Docker; local training runs use a pinned TAO container and require 8 GB of shared memory.
Requires the NVIDIA container toolkit alongside Docker.
Requires at least one GPU with 16 GB or more VRAM; eight GPUs are recommended for distributed training.
Access · 1
This skill
Hugging Face
Downloads backbone weights from Hugging Face