Sparse4D

Sparse4D for multi-camera temporal 3D object detection and tracking. Uses sparse queries with deformable attention across camera views and time for end-to-end 3D perception. Includes instance bank for temporal tracking.

Use a pretrained ResNet-101 backbone when one is available by setting train.pretrainedmodelpath. For local smoke validation, Sparse4D training can run with an empty train.pretrainedmodelpath, but production runs should still use a compatible PTM.

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 still requires schemas/train.schema.json and references/spectemplatetrain.yaml to exist and parse. Use the packaged train 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: sparse4d
• Formats: ovpkl
• Monitoring metric: val_mAP
• Current TAO Sparse4D training emits this value in status/logs as

imgbboxNuScenes/mAP and mAP; AutoML metric extractors should treat those emitted keys as aliases for valmAP. Multi-fidelity AutoML algorithms such as Hyperband, ASHA, and BOHB may promote a checkpoint to a resume job that completes without emitting a fresh valmAP alias. In that case, compare AutoML's carried metric to the source rung job that emitted imgbboxNuScenes/mAP or mAP, while still verifying that the promoted job resumed from the explicit epoch/step checkpoint, produced a real checkpoint, and is usable for evaluate/inference.

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"
CONVERTED_SCENE = "<scene-from-converter>"  # e.g. "subsetscene+bev-sensor-random-0"

train (mandatory data sources):

CONVERTED = "s3://bucket/results/<dataset_convert_job_id>"
{
    "train.num_epochs": 30,
    "train.checkpoint_interval": 10,
    "train.validation_interval": 10,
    "train.num_gpus": 1,
    "dataset.sequences.split_num": 90,
    "dataset.train_dataset.sequences_split_num": 90,
    "dataset.data_root": f"{S3_TRAIN}/train",
    "model.head.instance_bank.anchor": f"{CONVERTED}/anchor_init.npy",
    "dataset.train_dataset.ann_file": f"{CONVERTED}/train/{CONVERTED_SCENE}_infos_train.pkl",
    "dataset.val_dataset.ann_file": f"{CONVERTED}/val/{CONVERTED_SCENE}_infos_val.pkl",
    "dataset.test_dataset.ann_file": f"{CONVERTED}/test/{CONVERTED_SCENE}_infos_test.pkl",
}

evaluate (mandatory data sources):

CONVERTED = "s3://bucket/results/<dataset_convert_job_id>"
{
    "dataset.data_root": f"{S3_EVAL}/val",
    "model.head.instance_bank.anchor": f"{CONVERTED}/anchor_init.npy",
    "dataset.train_dataset.ann_file": f"{CONVERTED}/train/{CONVERTED_SCENE}_infos_train.pkl",
    "dataset.val_dataset.ann_file": f"{CONVERTED}/val/{CONVERTED_SCENE}_infos_val.pkl",
    "dataset.test_dataset.ann_file": f"{CONVERTED}/test/{CONVERTED_SCENE}_infos_test.pkl",
}

export (mandatory data sources):

CONVERTED = "s3://bucket/results/<dataset_convert_job_id>"
{
    "model.head.instance_bank.anchor": f"{CONVERTED}/anchor_init.npy",
}

inference (mandatory data sources):

CONVERTED = "s3://bucket/results/<dataset_convert_job_id>"
{
    "dataset.data_root": f"{S3_EVAL}/test",
    "model.head.instance_bank.anchor": f"{CONVERTED}/anchor_init.npy",
    "dataset.train_dataset.ann_file": f"{CONVERTED}/train/{CONVERTED_SCENE}_infos_train.pkl",
    "dataset.val_dataset.ann_file": f"{CONVERTED}/val/{CONVERTED_SCENE}_infos_val.pkl",
    "dataset.test_dataset.ann_file": f"{CONVERTED}/test/{CONVERTED_SCENE}_infos_test.pkl",
}

quantize (mandatory data sources):

CONVERTED = "s3://bucket/results/<dataset_convert_job_id>"
{
    "dataset.data_root": f"{S3_TRAIN}/train",
    "model.head.instance_bank.anchor": f"{CONVERTED}/anchor_init.npy",
    "dataset.train_dataset.ann_file": f"{CONVERTED}/train/{CONVERTED_SCENE}_infos_train.pkl",
    "dataset.val_dataset.ann_file": f"{CONVERTED}/val/{CONVERTED_SCENE}_infos_val.pkl",
    "dataset.test_dataset.ann_file": f"{CONVERTED}/test/{CONVERTED_SCENE}_infos_test.pkl",
    "dataset.quant_calibration_dataset.images_dir": f"{S3_TRAIN}",
}

See references/localdockerconversion.md for local-docker conversion roots and mounts, H5 depth-path normalization, converted annotation filenames, smoke-run maxnumcams/anchor contracts for export compatibility, and converted-artifact verification before train/evaluate/inference.

Eval Dataset

Optional. Val/test splits configured via dataset ann_file paths.

Important Parameters

• model.backbone: Backbone. Default resnet_101.
• model.neck.outchannels: FPN output channels. Default 256. numouts=4.
• model.input_shape: Input image shape [W, H]. Default [1408, 512].
• model.head.num_output: Number of detection output queries. Default 300.
• model.head.num_decoder: Number of decoder layers. Default 6.
• model.head.temporal: Enable temporal reasoning. Default True.
• model.head.instancebank.numanchor: Instance bank anchors. Default 900.
• model.head.instancebank.numtemp_instances: Temporal instance count. Default 600.
• model.depthbranch.lossweight: Depth supervision loss weight. Default 0.2.
• dataset.batch_size: Per-GPU batch size. Default 2.
• dataset.num_frames: Sequence length. Default 200.
• dataset.classes: Detection classes. Default [person, gr1t2, agilitydigit, novacarter]. numids=70 for tracking.
• train.optim.lr: Learning rate. Default 5e-5. imgbackbone lrmult=0.2.
• train.lr_scheduler: Cosine scheduler with linear warmup (500 iters, ratio 0.333).
• train.gradclip.maxnorm: Gradient clipping. Default 25.
• train.precision: Options: bf16, fp16, fp32. Default bf16.
• evaluate.metrics: Eval metrics. Default ["detection"]. Optional tracking evaluation.
• evaluate.tracking.enabled: Enable tracking evaluation. tracking_threshold=0.2.

Multi-GPU / Multi-Node

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

• Multi-GPU strategy: ddpfindunusedparameterstrue (no fsdp support)
• sync_batchnorm is always enabled (True)
• Iterations per epoch computed as: numframes numbevgroups / (numnodes numgpus * batchsize)
• Scaling: When increasing GPUs, effective batch size grows and iterations-per-epoch shrinks proportionally

Multi-node env vars (set by orchestrator): WORLDSIZE, NODERANK, MASTERADDR, MASTERPORT, NUMGPUPER_NODE.

Hardware

Minimum 2 GPU(s), recommended 8 GPU(s). 40GB+ (A100 recommended) VRAM per GPU. Multi-camera temporal model is memory intensive. bf16 required for practical training. Multi-GPU strongly recommended. Instance bank requires substantial memory for temporal reasoning.

Error Patterns

datasetconvert required: Must run datasetconvert first to produce annotation pickles and anchor_init.npy.

datasetconvert container/command: Sparse4D conversion is an AICity to OVPKL annotations conversion. Launch datasetconvert with the action-level taotoolkit.dataservices image and annotations convert -e {config_path}; do not use the PyTorch sparse4d CLI for conversion. Train/evaluate/export/ inference still use the model-level PyTorch image.

Stable raw-data path: The AICity to OVPKL converter writes image paths into the generated pickle files. Keep aicity.root at /data/aicityroot during conversion, then point dataset.dataroot at the split folder, for example /data/aicityroot/train for training or /data/aicityroot/val for evaluation. This preserves the converter's absolute RGB paths and relative depth paths.

H5 depth tuple mismatch: If training fails with an H5 path error where the trainer tries to open a camera directory such as /data/aicityroot/train/<scene>/Camera, run models/sparse4d/scripts/normalizedepthpaths.py --data-root <host-aicity-root>/train <converted-ann-dir> after datasetconvert and before train/evaluate/inference. The helper rewrites converted depthmappath tuples to point at <scene>/depth_maps/<camera>.h5 with the H5 dataset key basename.

Missing anchor file: Set model.head.instancebank.anchor to the anchorinit.npy path from dataset_convert results.

Temporal OOM: Reduce dataset.numframes or dataset.batchsize if running out of memory during temporal training.

Quantize image compatibility: The model-skill wiring should pass quantize.modelpath through the parent-model resolver, and checkpoint handoff should select the exact epoch/step checkpoint just like evaluate, inference, export, and resume. TorchAO checkpoint quantization passes in the validation-fixes-20260525 PyT image and writes quantizedmodel_torchao.pth. Older 7.0.0-rc PyT images may fail inside the Sparse4D quantize entrypoint or lack ONNX quantization dependencies; do not remove or skip the advertised quantize action if that occurs. Report the container/image failure and keep the exact checkpoint path visible.

Spec Param / Parent Model Inference

See references/specparaminference.md for the model-specific inference mappings from TAO Core sparse4d.config.json (the per-action spec-field to inference-function table) and the parentmodel/parentjobid checkpoint-resolution rules that generated runners apply with SDK helpers before createjob().