nicetoolbox.detectors.data¶
Data module handling the data loading and processing of the give datasets.
Classes
Facade for NICE Toolbox input data preparation. |
- class nicetoolbox.detectors.data.SequenceData(sequence_context: SequenceRuntimeConfig, io: SequenceIO)[source]¶
Facade for NICE Toolbox input data preparation.
Determines which modalities to prepare based on ‘input_data_type’ fields in the selected detector/algorithm configs from detectors_config.toml.
Supported input_data_type values: - “video” : video files / frame sequences (always prepared by default) - “audio” : audio tracks extracted from video or standalone files - “frames” : (future) non-temporal image datasets
- Algorithms may declare a single string or a list for cross-modal use:
input_data_type = “video” input_data_type = [“video”, “audio”]
Feature detectors (those with input_detector_names instead of input_data_type) do not trigger raw data preparation directly.
Initialize data facade and orchestrate data handling.
- property calibration: Dict[str, Any] | None¶
Camera calibration data (video-specific).
- get_audio_input_recipe() AudioInputRecipe | None[source]¶
Get audio input recipe for audio data loaders.
- get_input_recipes() InputRecipes[source]¶
Get composed InputRecipes containing all available modality recipes.
This is the primary method used by BaseMethod to build the runtime config for subprocesses. Each recipe is validated via Pydantic.
- Returns:
InputRecipes with video and/or audio recipes set.