physical-ai-video-data-augmentation

Orchestrate video data augmentation on NVIDIA OSMO by submitting a run with storage_url, dataset, run_id, and gpu_platform. The workflow manages the complete pipeline, deriving storage and enforcing preflight guard checks before execution.

Supported video augmentation flows include augmentation_and_al, auto_labeling, e2e, and e2e_super_resolution. These enable scalable video labeling and generate per-video configurations with deterministic seeds and variable sampling for diverse outputs.

Prevent failed OSMO video augmentation submissions by enforcing built-in preflight and pre-submit guard checks. The system validates dataset inputs, cache wiring, GPU pool availability, and credential provisioning before allowing any workflow to execute.

Run video augmentation without a full data backend by utilizing the provided demo datasets and the setup_model_cache path. This allows workflow execution in environments lacking complete storage infrastructure for testing and validation.

Start an end-to-end video augmentation run by providing a storage_url, dataset, run_id, video, and gpu_platform. The system derives the necessary storage URL from the dataset input and validates it during the preflight phase.

Video augmentation workflows require preflight guard checks to ensure storage and orchestration safety. Validating dataset URL inputs, cache wiring, GPU availability, and credentials before submission prevents resource allocation failures during execution.