Agent skill · nvidia
tao-mine-od-images
Run TAO Data Services TMM unique-neighbor matching mining from embedding parquet files for object detection workflows. Use when an object detection workflow needs to mine a bijectively-assigned set of unique source images closest to target samples. Use global allocation when mining without class constraints. Use class_stratified when rare classes are specified.
What it needs
About 4k tokens when loaded.
What this skill does
TAO Mine OD Images (Unique Neighbor Matching) Use this skill to run TAO Data Services TMM unique-neighbor matching mining for object detection. The skill consumes pre-embedded source and target parquets and writes a directory of outputs including finaluniquefiles.parquet and summary.json. It does not compute embeddings; upstream steps must produce the source and target embedding parquets first. The container entrypoint is: Inputs The user can provide either an existing spec or the fields needed to generate one. Required spec fields: Field Meaning --- --- sourcepath Absolute path to the source embeddings parquet or directory of parquets. targetpath Absolute path to the target embeddings parquet or directory of parquets. outputdir Absolute path to the output directory. Writes finaluniquefiles.parquet, summary.json, and per-iteration parquets. desireduniquecount Total number of unique source files to retrieve. Common optional fields: Field Default Meaning --- ---: --- allocationpolicy global global or classstratified. distancemetric euclidean One of euclidean, cosine, or manhattan. Embeddings are L2-normalized before search. candidateexpansionfactor 5 Candidate-pool multiplier per iteration. Increase if desired count is not reached. sourceembeddingcolumn embedding Embedding column in sourcepath. targetembeddingcolumn embedding Embedding column in targetpath. sourcefilepathcolumn filepath Filepath column in sourcepath; also the column of finaluniquefiles.parquet. targetfilepathcolumn filepath Filepath column in targetpath. excludepath null Parquet with a filepath column; those images are removed from the source pool. sourcedetectionfile null COCO .json or KITTI label directory for the source. Required for classstratified. targetdetectionfile null COCO .json or KITTI label directory for the target. Required for classstratified. detectionformat null coco or kitti. Required whenever a detection file is set; never inferred from the path. …
How to use it
Reference it in AdaL, Claude Code, Cursor or any coding agent — nothing to install:
@skills nvidia/tao-mine-od-images