Find fresh training images like your weak examples.
Selects source images similar to weak target examples to create a deduplicated training-augmentation queue.
When to use it
Use immediately after routing Visual ChangeNet samples when expanding the real-image augmentation queue from the mining subset.
Give it a target-image parquet and a source pool; it produces a deduplicated parquet of similar source-image paths.
What you provide
Requires Docker to pull and run the data-services container.
Requires the NVIDIA container toolkit for the GPU container workflow.
This skill
TAO data-services
Downloads the data-services container
alpine
An unpinned Alpine container