s6.app.cog.augmented_dataset¶
Augmented keypoint dataset and image transforms.
Wraps s6.utils.datapipe.StructuredDatasetTorch to produce square
patches around configured image/keypoint pairs with optional segmentation masks.
One raw multi-camera record can expand into multiple trainable samples through
data_mappings.
- class s6.app.cog.augmented_dataset.AugmentationConfigBase(*, name: str, enabled: bool = True)
Bases:
BaseModel- name: str
- enabled: bool
- class s6.app.cog.augmented_dataset.MirrorAugmentationConfig(*, name: Literal['mirror'] = 'mirror', enabled: bool = False)
Bases:
AugmentationConfigBase- name: Literal['mirror']
- enabled: bool
- class s6.app.cog.augmented_dataset.CenterSquareCropAugmentationConfig(*, name: Literal['center_square_crop'] = 'center_square_crop', enabled: bool = True)
Bases:
AugmentationConfigBase- name: Literal['center_square_crop']
- enabled: bool
- class s6.app.cog.augmented_dataset.CropAugmentationConfig(*, name: Literal['crop'] = 'crop', enabled: bool = True, crop_factor: float | Tuple[float, float] = 0.9, output_size: int = 256, max_rotation: float = 0.0)
Bases:
AugmentationConfigBase- name: Literal['crop']
- enabled: bool
- crop_factor: float | Tuple[float, float]
- output_size: int
- max_rotation: float
- class s6.app.cog.augmented_dataset.RotationAugmentationConfig(*, name: Literal['rotation'] = 'rotation', enabled: bool = True, max_rotation: float = 360.0)
Bases:
AugmentationConfigBase- name: Literal['rotation']
- enabled: bool
- max_rotation: float
- class s6.app.cog.augmented_dataset.BlurAugmentationConfig(*, name: Literal['blur'] = 'blur', enabled: bool = False, sigma_range: Tuple[float, float] = (0.0, 2.0))
Bases:
AugmentationConfigBase- name: Literal['blur']
- enabled: bool
- sigma_range: Tuple[float, float]
- class s6.app.cog.augmented_dataset.OcclusionAugmentationConfig(*, name: Literal['occlusion'] = 'occlusion', enabled: bool = False, size_range: Tuple[float, float] = (0.1, 0.3), intensity_range: Tuple[int, int] = (0, 255))
Bases:
AugmentationConfigBase- name: Literal['occlusion']
- enabled: bool
- size_range: Tuple[float, float]
- intensity_range: Tuple[int, int]
- class s6.app.cog.augmented_dataset.ColorJitterAugmentationConfig(*, name: Literal['color_jitter'] = 'color_jitter', enabled: bool = True, brightness: float = 0.2, contrast: float = 0.2, gamma: float = 0.2)
Bases:
AugmentationConfigBase- name: Literal['color_jitter']
- enabled: bool
- brightness: float
- contrast: float
- gamma: float
- class s6.app.cog.augmented_dataset.SamplingConfig(*, enabled: bool = False, bins: int = 10, seed: int = 0)
Bases:
BaseModelConfigure distance-based bucket sampling of keypoints.
- enabled: bool
- bins: int
- seed: int
- class s6.app.cog.augmented_dataset.DataMappings(*, x: List[str] = None, y: List[str] = None, mask: List[str] | None = None)
Bases:
BaseModelLogical training-target mappings to raw StructuredDataset datakeys.
- x: List[str]
- y: List[str]
- mask: List[str] | None
- property has_masks: bool
- raw_datakeys() List[str]
- class s6.app.cog.augmented_dataset.TrainingMappingRef(image_key: str, keypoint_key: str, mask_key: str | None = None)
Bases:
object- image_key: str
- keypoint_key: str
- mask_key: str | None = None
- class s6.app.cog.augmented_dataset.SampleRef(dataset_idx: int, source_idx: int, mapping_idx: int)
Bases:
object- dataset_idx: int
- source_idx: int
- mapping_idx: int
- class s6.app.cog.augmented_dataset.AugmentedKeypointDataset(config: Config)
Bases:
DatasetDataset that applies keypoint-centric augmentations.
- class Config(*, prefix: ConstrainedStrValue = 'model', base_dir: str | List[str] = [], data_mappings: DataMappings = None, augmentations: List[dict] = None, sd_kwargs: dict = {}, sampling: SamplingConfig = SamplingConfig(enabled=False, bins=10, seed=0), num_segmentation_classes: ConstrainedIntValue | None = None, segmentation_loss_weight: ConstrainedFloatValue = 1.0)
Bases:
BaseModel- prefix: str
- base_dir: str | List[str]
- data_mappings: DataMappings
- augmentations: List[dict]
- sd_kwargs: dict
- sampling: SamplingConfig
- num_segmentation_classes: int | None
- segmentation_loss_weight: float
- classmethod default()
- classmethod from_json(cfg_path: str) AugmentedKeypointDataset
- property output_size: int
- property has_segmentation: bool
- s6.app.cog.augmented_dataset.center_square_crop(image: ndarray, keypoint: Tuple[int, int], mask: ndarray | None = None) Tuple[ndarray, ndarray, ndarray | None]
- s6.app.cog.augmented_dataset.random_crop_and_resize(image: ndarray, keypoint: Tuple[int, int], crop_factor: float | Tuple[float, float] = 0.8, output_size: int = 256, max_rotation: float = 0.0, mask: ndarray | None = None) Tuple[ndarray, ndarray, ndarray | None]
- s6.app.cog.augmented_dataset.random_rotate(image: ndarray, keypoint: ndarray, max_angle: float = 30.0, mask: ndarray | None = None) Tuple[ndarray, ndarray, ndarray | None]
- s6.app.cog.augmented_dataset.random_color_jitter(image: ndarray, brightness: float = 0.2, contrast: float = 0.2, gamma: float = 0.2) ndarray
- s6.app.cog.augmented_dataset.horizontal_mirror(image: ndarray, keypoint: Tuple[int, int], mask: ndarray | None = None) Tuple[ndarray, ndarray, ndarray | None]
- s6.app.cog.augmented_dataset.random_blur(image: ndarray, sigma_range: Tuple[float, float]) ndarray
- s6.app.cog.augmented_dataset.random_occlusion(image: ndarray, keypoint: ndarray, size_range: Tuple[float, float], intensity_range: Tuple[int, int], mask: ndarray | None = None) Tuple[ndarray, ndarray | None]