Action Recognition

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).

Action recognition from video sequences. Supports RGB, optical flow, and joint (multi-stream) input types for classifying temporal actions in video clips.

Set model.pretrainedmodelpath for pretrained backbone weights.

Quick Start (docker run)

Docker-native launch — no TAO SDK and no Python on the host. Use the local Docker/platform skill instead when it gives a stricter environment-specific command (non-root UID mapping, cache redirects, remote daemons).

TAO_PYT_IMAGE_DEFAULT=nvcr.io/nvidia/tao/tao-toolkit:7.1.0-pyt  # versions-key: images.tao_toolkit.pyt
TAO_PYT_IMAGE="${TAO_PYT_IMAGE:-$TAO_PYT_IMAGE_DEFAULT}"
RUN_ROOT="${RUN_ROOT:-$PWD}"
DOCKER_COMMON=(
  --rm --gpus all --shm-size=8g
  --shm-size=8g
  --ulimit memlock=-1
  --ulimit stack=67108864
  -v "$RUN_ROOT/data:/data:ro"
  -v "$RUN_ROOT/specs:/specs:ro"
  -v "$RUN_ROOT/results:/results"
)

Train:

docker run "${DOCKER_COMMON[@]}" "$TAO_PYT_IMAGE" \
  action_recognition train -e /specs/train.yaml

Evaluate:

docker run "${DOCKER_COMMON[@]}" "$TAO_PYT_IMAGE" \
  action_recognition evaluate -e /specs/evaluate.yaml

Inference:

docker run "${DOCKER_COMMON[@]}" "$TAO_PYT_IMAGE" \
  action_recognition inference -e /specs/inference.yaml

Export:

docker run "${DOCKER_COMMON[@]}" "$TAO_PYT_IMAGE" \
  action_recognition export -e /specs/export.yaml

Every action takes its spec with -e; results_dir is set in the spec or overridden on the command line. Mount any pretrained-weights directory the spec references, and keep every in-container path consistent across actions.

Dataclass Schemas

Generated TAO Core schemas are packaged in schemas/<action>.schema.json, with schemas/manifest.json listing available actions. Each generated schema also emits references/spectemplate<action>.yaml from the schema top-level default field. AutoML enablement is declared at the model layer in references/skillinfo.yaml via automlenabled. Runnable AutoML for an action requires schemas/<action>.schema.json and references/spectemplate<action>.yaml to exist and parse. Use the packaged selected-action schema for automldefaultparameters, automldisabledparameters, defaults, min/max bounds, enums, option weights, math conditions, dependencies, and popular parameters. Do not expect ~/tao-core at runtime; maintainers regenerate schemas/templates before packaging the skill bank.

Train Action Policy

This model is AutoML-enabled at the model layer. Before handling any train-stage request, read references/skillinfo.yaml and resolve the run override from either an explicit automlpolicy value or the user's workflow request. Use automlpolicy: on by default and only expose on / off in new launch prompts. Treat phrases like "turn off AutoML", "disable AutoML", "no HPO", or "plain training" as automlpolicy: off for this run only. When automlpolicy: on, automlenabled: true, and both schemas/train.schema.json and references/spectemplatetrain.yaml are packaged, route the train action through tao-skill-bank:tao-run-automl by default with this model's skilldir. Preserve workflow/application overrides for datasets, specs, output directories, GPU/platform settings, parent checkpoints, and automlpolicy. Use direct model training only when automl_policy: off or the packaged train schema/template is missing; in the missing-schema case, report that AutoML is enabled but not runnable for this model until schemas are generated.

Non-train actions such as evaluate, inference, export, and deploy flows stay in this model skill. The per-run automl_policy override does not change model metadata.

Training Requirements

• Dataset type: action_recognition
• Formats: default
• Monitoring metric: val_loss

Per-Action Dataset Requirements

Typical Spec Overrides

Data source overrides are mandatory for every action — the agent MUST construct data source paths from the Per-Action Dataset Requirements table above and include them in spec_overrides.

LOCAL_DATA = "/workspace/data/extracted"

If the source dataset is provided as the TAO sample archives train.tar.gz, test.tar.gz, or test/smile.tar.gz, download and extract them before launching the TAO container. The action-recognition entrypoints expect directory paths and fail with NotADirectoryError when these spec keys point at .tar.gz files.

train (mandatory data sources):

{
    "train.num_epochs": 30,
    "train.checkpoint_interval": 10,
    "train.validation_interval": 10,
    "train.num_gpus": 1,
    "dataset.label_map": {
        "catch": 0,
        "smile": 1
    },
    "dataset.batch_size": 2,
    "dataset.train_dataset_dir": f"{LOCAL_DATA}/train",
    "dataset.val_dataset_dir": f"{LOCAL_DATA}/test",
}

evaluate (mandatory data sources):

{
    "dataset.label_map": {
        "catch": 0,
        "smile": 1
    },
    "evaluate.test_dataset_dir": f"{LOCAL_DATA}/test",
}

inference (mandatory data sources):

{
    "dataset.label_map": {
        "catch": 0,
        "smile": 1
    },
    "inference.inference_dataset_dir": f"{LOCAL_DATA}/smile_infer/smile",
}

export (mandatory checkpoint + output path):

{
    "export.checkpoint": "<selected train checkpoint>",
    "export.onnx_file": "<results_dir>/action_recognition.onnx",
}

For direct local-docker chaining without the SDK resolver, select the concrete checkpoint produced by training, for example modelepoch000step00005.pth, and pass that exact file to evaluate, inference, and export. Do not use the armodellatest.pth symlink unless the user explicitly requests latest-checkpoint behavior. For resume training, set train.resumetrainingcheckpoint_path to the exact epoch/step checkpoint being resumed.

Eval Dataset

Optional. Test dataset may be distributed as test.tar.gz separate from training; extract it and point the spec to the extracted test/ directory. TAO training emits valloss as the validation scalar for the packaged sample data; use valloss with minimize direction for AutoML selection unless a custom evaluator supplies an accuracy metric.

Important Parameters

• model.model_type: Input type: rgb, of (optical flow), or joint (multi-stream).
• model.backbone: Default resnet_18. Used as the spatial feature extractor.
• dataset.label_map: Dictionary mapping class names to indices.
• model.rgbseqlength: Number of frames per clip for RGB input.
• model.ofseqlength: Number of frames for optical flow input.
• train.optim.lr: Learning rate. Default 5e-4.

Multi-GPU / Multi-Node

Launch method: Lightning-managed (single python process, Lightning spawns workers).

• Strategy: auto (Lightning picks best strategy automatically)
• No explicit numnodes or distributedstrategy config — single-node oriented

Hardware

Minimum 1 GPU(s), recommended 2 GPU(s). 16GB+ VRAM per GPU. Memory depends on sequence length and input resolution. batch_size=2 is conservative for video data.

Error Patterns

Sequence length mismatch: Ensure video clips have enough frames for the configured rgbseqlength or ofseqlength.

Evaluate/inference missing label map: Downstream actions rebuild the ActionRecognitionModel before loading the checkpoint, so they need the same dataset.label_map used during training. Include it with every evaluate or inference spec; otherwise model construction fails before the checkpoint can be validated.

Spec Param / Parent Model Inference

Model-specific inference mappings belong in this MD file, not in config.json. Generated runners should read this section and apply the mappings with SDK helpers before createjob(). This mirrors the old microservices inferparams.py flow.

Inference mappings from TAO Core action_recognition.config.json:

For parentmodel or parentmodelfolder, pass the upstream train/export/AutoML child job id as parentjob_id. The SDK lists the parent result folder, filters checkpoint artifacts, and returns the selected model file or folder. Do not add these mappings back to config.json and do not patch generated runner scripts to guess checkpoint paths.