Segmentation Train Config

Bases: TrainConfig

Training configuration for instance segmentation models.

Extends :class:TrainConfig with segmentation-specific loss coefficients.

Attributes:

Name	Type	Description
num_select	Optional[int]	Maximum number of predictions to keep per image. None uses the model default.
mask_point_sample_ratio	int	Number of points sampled per mask for point-based mask loss computation.
mask_ce_loss_coef	float	Cross-entropy loss weight for mask prediction.
mask_dice_loss_coef	float	Dice loss weight for mask prediction.
cls_loss_coef	float	Classification loss weight. Defaults to 1.0 to match the effective pre-v1.7 value (the v1.7 TrainConfig ownership migration silently activated a dormant 5.0; this field restores the correct weight). To reproduce pre-fix segmentation behaviour pass cls_loss_coef=5.0 explicitly.
segmentation_head	bool	Whether to attach the segmentation head.

Source code in src/rfdetr/config.py

class SegmentationTrainConfig(TrainConfig):
    """Training configuration for instance segmentation models.

    Extends :class:`TrainConfig` with segmentation-specific loss coefficients.

    Attributes:
        num_select: Maximum number of predictions to keep per image. ``None`` uses
            the model default.
        mask_point_sample_ratio: Number of points sampled per mask for point-based
            mask loss computation.
        mask_ce_loss_coef: Cross-entropy loss weight for mask prediction.
        mask_dice_loss_coef: Dice loss weight for mask prediction.
        cls_loss_coef: Classification loss weight. Defaults to ``1.0`` to match the
            effective pre-v1.7 value (the v1.7 TrainConfig ownership migration
            silently activated a dormant ``5.0``; this field restores the correct
            weight). To reproduce pre-fix segmentation behaviour pass
            ``cls_loss_coef=5.0`` explicitly.
        segmentation_head: Whether to attach the segmentation head.
    """

    num_select: Optional[int] = None
    mask_point_sample_ratio: int = 16
    mask_ce_loss_coef: float = 5.0
    mask_dice_loss_coef: float = 5.0
    cls_loss_coef: float = 1.0
    segmentation_head: bool = True

Functions

expand_paths(v) classmethod

Expand and normalize dataset/output directory paths via os.fspath → expanduser → realpath.

Source code in src/rfdetr/config.py

@field_validator("dataset_dir", "output_dir", mode="before")
@classmethod
def expand_paths(cls, v: PathLikeStr | None) -> str | None:
    """Expand and normalize dataset/output directory paths via ``os.fspath`` → ``expanduser`` → ``realpath``."""
    if v is None:
        return v
    return os.path.realpath(os.path.expanduser(os.fspath(v)))

validate_batch_size(v) classmethod

Validate batch_size is a positive integer or the literal 'auto'.

Source code in src/rfdetr/config.py

@field_validator("batch_size", mode="after")
@classmethod
def validate_batch_size(cls, v: int | Literal["auto"]) -> int | Literal["auto"]:
    """Validate batch_size is a positive integer or the literal 'auto'."""
    if v == "auto":
        return v
    if v < 1:
        raise ValueError("batch_size must be >= 1, or 'auto'.")
    return v

validate_ema_headroom(v) classmethod

Validate auto_batch_ema_headroom is in (0, 1].

Source code in src/rfdetr/config.py

@field_validator("auto_batch_ema_headroom", mode="after")
@classmethod
def validate_ema_headroom(cls, v: float) -> float:
    """Validate auto_batch_ema_headroom is in (0, 1]."""
    if not (0 < v <= 1.0):
        raise ValueError("auto_batch_ema_headroom must be in (0, 1].")
    return v

validate_positive_intervals(v) classmethod

Validate interval fields are >= 1.

Source code in src/rfdetr/config.py

@field_validator("ema_update_interval", "eval_interval", mode="after")
@classmethod
def validate_positive_intervals(cls, v: int) -> int:
    """Validate interval fields are >= 1."""
    if v < 1:
        raise ValueError("Interval fields must be >= 1.")
    return v

validate_positive_train_steps(v) classmethod

Validate accumulation, target-effective batch, and max targets are >= 1.

Source code in src/rfdetr/config.py

@field_validator(
    "grad_accum_steps", "auto_batch_target_effective", "auto_batch_max_targets_per_image", mode="after"
)
@classmethod
def validate_positive_train_steps(cls, v: int) -> int:
    """Validate accumulation, target-effective batch, and max targets are >= 1."""
    if v < 1:
        raise ValueError(
            "grad_accum_steps, auto_batch_target_effective, and auto_batch_max_targets_per_image must be >= 1."
        )
    return v

validate_prefetch_factor(v) classmethod

Validate prefetch_factor is None or >= 1.

Source code in src/rfdetr/config.py

@field_validator("prefetch_factor", mode="after")
@classmethod
def validate_prefetch_factor(cls, v: Optional[int]) -> Optional[int]:
    """Validate prefetch_factor is None or >= 1."""
    if v is not None and v < 1:
        raise ValueError("prefetch_factor must be >= 1 when provided.")
    return v

validate_smooth_alpha(v) classmethod

Validate smooth_alpha is in [0.0, 1.0).

Source code in src/rfdetr/config.py

@field_validator("smooth_alpha", mode="after")
@classmethod
def validate_smooth_alpha(cls, v: float) -> float:
    """Validate smooth_alpha is in [0.0, 1.0)."""
    if not (0.0 <= v < 1.0):
        raise ValueError("smooth_alpha must be in [0.0, 1.0).")
    return v