Tao Train Mask2former

Mask2Former

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).

Mask2Former for universal image segmentation (panoptic, instance, and semantic). Transformer-based with masked attention for high-quality segmentation results.

Set model.backbone.pretrained_weights for Swin backbone weights.

For TAO Deploy TensorRT actions (gen_trt_engine, TensorRT evaluate, and TensorRT inference), read references/tao-deploy-mask2former.md first. Deploy spec templates live in this skill's references/ folder with the spec_template_deploy_*.yaml prefix.

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/spec_template_<action>.yaml from the schema top-level default field. AutoML enablement is declared at the model layer in references/skill_info.yaml via automl_enabled. Runnable AutoML for an action requires schemas/<action>.schema.json and references/spec_template_<action>.yaml to exist and parse. Use the packaged selected-action schema for automl_default_parameters, automl_disabled_parameters, 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/skill_info.yaml and resolve the run override from either an explicit automl_policy value or the user's workflow request. Use automl_policy: 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 automl_policy: off for this run only. When automl_policy: on, automl_enabled: true, and both schemas/train.schema.json and references/spec_template_train.yaml are packaged, route the train action through tao-skill-bank:tao-run-automl by default with this model's skill_dir. Preserve workflow/application overrides for datasets, specs, output directories, GPU/platform settings, parent checkpoints, and automl_policy. 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: segmentation
• Formats: coco_panoptic, coco
• Monitoring metric: mIoU
• AutoML metric: Maximize emitted validation/evaluation mIoU; never substitute training 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.

S3_TRAIN = "s3://bucket/data/train"
S3_EVAL = "s3://bucket/data/eval"

train (mandatory data sources):

{
    "train.num_gpus": 1,
    "train.num_epochs": 10,
    "train.checkpoint_interval": 10,
    "train.validation_interval": 10,
    "model.sem_seg_head.num_classes": 133,
    "dataset.contiguous_id": True,
    "dataset.train.type": "coco_panoptic",
    "dataset.val.type": "coco_panoptic",
    "dataset.test.type": "coco_panoptic",
    "dataset.train.img_dir": f"{S3_TRAIN}/images.tar.gz",
    "dataset.label_map": f"{S3_TRAIN}/label_map_panoptic.json",
    "dataset.train.instance_json": f"{S3_TRAIN}/annotations.json",
    "dataset.train.panoptic_json": f"{S3_TRAIN}/annotations_panoptic.json",
    "dataset.train.panoptic_dir": f"{S3_TRAIN}/images_panoptic.tar.gz",
    "dataset.val.img_dir": f"{S3_EVAL}/images.tar.gz",
    "dataset.val.instance_json": f"{S3_EVAL}/annotations.json",
    "dataset.val.panoptic_json": f"{S3_EVAL}/annotations_panoptic.json",
    "dataset.val.panoptic_dir": f"{S3_EVAL}/images_panoptic.tar.gz",
    "dataset.test.img_dir": f"{S3_EVAL}/images.tar.gz",
}

evaluate (mandatory data sources):

{
    "evaluate.checkpoint": "<selected train/AutoML checkpoint>",
    "model.sem_seg_head.num_classes": 133,
    "dataset.contiguous_id": True,
    "dataset.train.type": "coco_panoptic",
    "dataset.val.type": "coco_panoptic",
    "dataset.test.type": "coco_panoptic",
    "dataset.train.img_dir": f"{S3_TRAIN}/images.tar.gz",
    "dataset.label_map": f"{S3_TRAIN}/label_map_panoptic.json",
    "dataset.train.instance_json": f"{S3_TRAIN}/annotations.json",
    "dataset.train.panoptic_json": f"{S3_TRAIN}/annotations_panoptic.json",
    "dataset.train.panoptic_dir": f"{S3_TRAIN}/images_panoptic.tar.gz",
    "dataset.val.img_dir": f"{S3_EVAL}/images.tar.gz",
    "dataset.val.instance_json": f"{S3_EVAL}/annotations.json",
    "dataset.val.panoptic_json": f"{S3_EVAL}/annotations_panoptic.json",
    "dataset.val.panoptic_dir": f"{S3_EVAL}/images_panoptic.tar.gz",
    "dataset.test.img_dir": f"{S3_EVAL}/images.tar.gz",
}

export:

{
    "export.checkpoint": "<selected train/AutoML checkpoint>",
    "export.onnx_file": "<output ONNX path>",
    "model.sem_seg_head.num_classes": "<same value used for train>",
}

inference (mandatory data sources):

{
    "inference.checkpoint": "<selected train/AutoML checkpoint>",
    "model.sem_seg_head.num_classes": "<same value used for train>",
    "dataset.contiguous_id": True,
    "dataset.train.img_dir": f"{S3_TRAIN}/images.tar.gz",
    "dataset.label_map": f"{S3_TRAIN}/label_map_panoptic.json",
    "dataset.train.instance_json": f"{S3_TRAIN}/annotations.json",
    "dataset.train.panoptic_json": f"{S3_TRAIN}/annotations_panoptic.json",
    "dataset.train.panoptic_dir": f"{S3_TRAIN}/images_panoptic.tar.gz",
    "dataset.val.img_dir": f"{S3_EVAL}/images.tar.gz",
    "dataset.val.instance_json": f"{S3_EVAL}/annotations.json",
    "dataset.val.panoptic_json": f"{S3_EVAL}/annotations_panoptic.json",
    "dataset.val.panoptic_dir": f"{S3_EVAL}/images_panoptic.tar.gz",
    "dataset.test.img_dir": f"{S3_EVAL}/images.tar.gz",
}

quantize (mandatory data sources):

{
    "quantize.model_path": "<selected train/export artifact>",
    "dataset.train.img_dir": f"{S3_TRAIN}/images.tar.gz",
    "dataset.label_map": f"{S3_TRAIN}/label_map_panoptic.json",
    "dataset.train.instance_json": f"{S3_TRAIN}/annotations.json",
    "dataset.train.panoptic_json": f"{S3_TRAIN}/annotations_panoptic.json",
    "dataset.train.panoptic_dir": f"{S3_TRAIN}/images_panoptic.tar.gz",
    "dataset.val.img_dir": f"{S3_EVAL}/images.tar.gz",
    "dataset.val.instance_json": f"{S3_EVAL}/annotations.json",
    "dataset.val.panoptic_json": f"{S3_EVAL}/annotations_panoptic.json",
    "dataset.val.panoptic_dir": f"{S3_EVAL}/images_panoptic.tar.gz",
    "dataset.test.img_dir": f"{S3_EVAL}/images.tar.gz",
    "dataset.quant_calibration_dataset.images_dir": f"{S3_TRAIN}/images.tar.gz",
}

Eval Dataset

Optional. Val data sources are part of the dataset config alongside train.

Important Parameters

• model.sem_seg_head.num_classes: Number of segmentation classes. Default 200. Must match your annotation categories.
• model.backbone.swin.type: Swin Transformer variant. Default tiny. Options include tiny, small, base, large.
• model.mode: Segmentation mode. Default panoptic. Options: panoptic, instance, semantic.
• train.optim.lr: Learning rate. Default 2e-4 (AdamW).
• dataset.train.batch_size: Per-GPU batch size. Default 1. Mask2Former is memory-intensive due to per-pixel predictions.
• dataset.contiguous_id: If true, set model.sem_seg_head.num_classes to the number of label-map categories. If false, set model.sem_seg_head.num_classes above the maximum raw category id and keep the same setting for evaluate, inference, export, deploy, and quantize. The COCO panoptic S3 sample has 133 categories with raw ids up to 200, so raw-id validation uses num_classes: 201.

Multi-GPU / Multi-Node

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

• Same DDP/FSDP behavior as DINO (activation checkpoint aware)
• FAN backbones auto-enable sync_batchnorm
• fsdp forces FP16

Multi-node env vars (set by orchestrator): WORLD_SIZE, NODE_RANK, MASTER_ADDR, MASTER_PORT, NUM_GPU_PER_NODE.