^j!XdZddlmZddlZddgZd dZd dZd dZd d Z dd Z y)zAKeypoint utility functions shared by inference and visualization.) annotationsNschemas_semantically_equal&precision_cholesky_to_pixel_covariancec,t|xr|ddk(S)aReturn True if *schema* uses a background-first layout. A background-first schema has a leading slot with zero keypoints that acts as a background/no-keypoint sentinel, e.g. ``[0, 17]`` where slot 0 is background and slot 1 is person with 17 keypoints. Active-first schemas omit that slot entirely, e.g. ``[17]``. Args: schema: Keypoints-per-class list. Returns: ``True`` when ``schema`` is non-empty and its first element is zero. Examples: >>> _is_bg_first_schema([0, 17]) True >>> _is_bg_first_schema([17]) False >>> _is_bg_first_schema([]) False r)boolschemas e/var/www/ramen.bs-engineer-server.com/venv/lib/python3.12/site-packages/rfdetr/utilities/keypoints.py_is_bg_first_schemar s, < *F1IN*c8t|r|ddSt|S)aStrip the leading background slot from a bg-first schema. Always returns a new list. A no-op (copy) when *schema* is already active-first or empty. Only the first leading zero slot is removed; schemas with multiple leading zeros (e.g. ``[0, 0, 17]``) retain all but the first. Args: schema: Keypoints-per-class list. Returns: Schema with the leading zero-keypoint slot removed when bg-first, or a copy of *schema* otherwise. Examples: >>> _to_active_first([0, 17]) [17] >>> _to_active_first([17]) [17] >>> _to_active_first([0, 17, 4]) [17, 4] >>> _to_active_first([0]) [] Nr listrs r _to_active_firstr+s"26"abz <r cPt|s|s t|Sdgt|zS)aPrepend a background slot to an active-first schema. A no-op when *schema* already starts with a zero-keypoint slot or is empty. Args: schema: Keypoints-per-class list. Returns: Schema with a leading ``0`` prepended when not already bg-first. Examples: >>> _to_bg_first([17]) [0, 17] >>> _to_bg_first([0, 17]) [0, 17] >>> _to_bg_first([17, 4]) [0, 17, 4] rrrs r _to_bg_firstrIs)&6"&F| 3f r c0t|t|k(S)asReturn True if *a* and *b* represent the same keypoint structure. Two schemas are semantically equal when they encode identical active keypoint counts per class after stripping any leading background slot. This allows ``[0, 17]`` (bg-first) and ``[17]`` (active-first) to compare equal. Args: a: First keypoints-per-class list. b: Second keypoints-per-class list. Returns: ``True`` when both schemas reduce to the same active-first form. Note: A schema of ``[0]`` (one detection-only background slot) is semantically equal to ``[]`` (no schema) because both reduce to an empty active-first form. Callers that need to distinguish "no schema" from "bg-first with no active slots" should compare ``to_active_first(a)`` directly or check ``bool(schema)`` before calling this function. Use this function for user-facing validation where representational form does not matter. Use exact ``!=`` comparison when preserving representational form is required (e.g. auto-align in checkpoint loading). Examples: >>> schemas_semantically_equal([0, 17], [17]) True >>> schemas_semantically_equal([17], [17]) True >>> schemas_semantically_equal([0, 17], [0, 33]) False >>> schemas_semantically_equal([0], []) True )r)abs r rrasH A "21"5 55r c.|jdk7s|jddk7rtd|jd|j|jddfk7r)td|jdd|jd|jtj d }|d }|d }|d }t j |jd }t j|}t j|}t jddd5d||z|z|zz } | ||z||zzz} | | |zzdz} | ||zz} |ddddfjtj } |ddddfjtj }||z| z}|| z| z}| | z| z}ddd|t j zt j zt j z}t jtj}t jg|jdddd|tj}t j||jtj||d<t j||jtj||d<|d|d<t j||jtj||d<|S#1swYZxYw)a.Convert RF-DETR keypoint precision parameters into pixel covariances. The keypoint head predicts lower-triangular precision-Cholesky parameters in normalized image coordinates. This helper inverts those precision matrices and scales them to pixel coordinates for Supervision covariance annotators. Args: precision_cholesky: Lower-triangular precision parameters with shape ``(N, K, 3)``. Each triplet is ``(log_l11, l21, log_l22)``. source_shape: Per-detection ``(height, width)`` rows with shape ``(N, 2)``. Returns: Pixel-space covariance matrices with shape ``(N, K, 2, 2)``. Raises: ValueError: If ``precision_cholesky`` or ``source_shape`` has an incompatible shape. Example: >>> precision = np.array([[[0.0, 0.0, 0.0]]], dtype=np.float32) >>> shape = np.array([[10.0, 20.0]], dtype=np.float32) >>> precision_cholesky_to_pixel_covariance(precision, shape)[0, 0] array([[400., 0.], [ 0., 100.]], dtype=float32) z2precision_cholesky must have shape (N, K, 3), got .rzsource_shape must have shape (z , 2), got F)copy).r).r).r)axisignore)divideinvalidoverg?gNr)dtype).rr).rr).rr).rr)ndimshape ValueErrorastypenpfloat64isfiniteallexperrstatefloat32nanfullwhere)precision_cholesky source_shapeprecision_cholesky_f64log_l11l21log_l22 finite_inputl11l22inv_detcov00cov01cov11heightwidthpx00px01px11 finite_allr. covariancess r rrs<!#'9'?'?'Ba'GMN`NfNfMgghijj066q91==9:L:R:RST:U9VV`amasas`ttuvww(066rzz6N$V,G  (C$V,G;;56:::CL &&/C &&/C HhX F'sS3./39sSy01C4#:&,39%a1f%,,RZZ8Q!V$++BJJ7u}u$v~%&'" D 11BKK4EE TXHYYJ **RVV C''?.44Ra8?!?Q?BJJWKXXj$++bjj2I3OK XXj$++bjj2I3OK (3K XXj$++bjj2I3OK  3''s BL  L)r list[int]returnr)r rErFrE)rrErrErFr)r1 np.ndarrayr2rGrFrG) __doc__ __future__rnumpyr'__all__r rrrrr r rMsX H"!,  +2<0$6N["[[[r