util/config.py

"""Centralized constants for the VANTAGE-Bench leaderboard.

All ordering, naming, and bucket definitions live here so that downstream
modules (data, ranking, render) read from a single source of truth.
"""

from __future__ import annotations


# -- Version / metadata ----------------------------------------------------

VERSION = "v1.0"
UPDATED = "2026-05-23"


# -- Schema -----------------------------------------------------------------

SCHEMA_VERSION = "1.0"


# -- Pillars ----------------------------------------------------------------
# Maps pillar key → ordered list of short task keys belonging to that pillar.
# 'overall' contains every task (used for averaging across all pillars).
# Short task keys match the TASKS dict below.

PILLARS: dict[str, list[str]] = {
    "overall":   ["loc", "ground", "pointing", "sot", "temploc", "dvc", "ev", "vqa"],
    "spatial":   ["loc", "ground", "pointing"],
    "st":        ["sot"],
    "temporal":  ["temploc", "dvc"],
    "semantic":  ["ev", "vqa"],
}


# -- Tasks ------------------------------------------------------------------
# Short task key → display label shown in column headers and side-panel.

TASKS: dict[str, str] = {
    "loc":      "2D Object Localization",
    "ground":   "2D Referring Expressions",
    "pointing": "2D Pointing",
    "sot":      "Single Object Tracking",
    "temploc":  "Temporal Localization",
    "dvc":      "Dense Video Captioning",
    "ev":       "Event Verification",
    "vqa":      "Video QA",
}


# -- Parameter buckets ------------------------------------------------------
# Short bucket key → display label for the filter UI.
# Matching logic lives in data.py (_param_bucket helper).

PARAM_BUCKETS: dict[str, str] = {
    "all":    "All sizes",
    "lt10":   "<10B",
    "10to40": "10B–40B",
    "gt40":   ">40B",
}


# -- Badge colors -----------------------------------------------------------
# badge type → (background_hex, text_hex, border_hex)

BADGE_COLORS: dict[str, tuple[str, str, str]] = {
    "verified":    ("#dcfce7", "#15803d", "#86efac"),
    "ensemble":    ("#dbeafe", "#1d4ed8", "#93c5fd"),
    "single":      ("#f3f4f6", "#374151", "#d1d5db"),
    "open":        ("#ccfbf1", "#0f766e", "#5eead4"),
    "proprietary": ("#fef3c7", "#92400e", "#fcd34d"),
    "new":         ("#ede9fe", "#5b21b6", "#c4b5fd"),
}


# ── Everything below this line is preserved for compatibility with existing
#    modules (data.py, ranking.py, render.py, filters.py, app.py).
#    Do not remove or rename these without updating the importing modules.
# --------------------------------------------------------------------------

# -- Canonical pillar + tab order ------------------------------------------

LEADERBOARD_TABS: list[str] = [
    "overall",
    "spatial",
    "spatio_temporal",
    "temporal",
    "semantic",
]

HEADLINE_FIELD_BY_TAB: dict[str, str] = {
    "overall": "overall",
    "spatial": "spatial",
    "spatio_temporal": "spatio_temporal",
    "temporal": "temporal",
    "semantic": "semantic",
}


# -- Required / optional field sets ----------------------------------------

REQUIRED_MODEL_FIELDS: list[str] = [
    "id",
    "name",
    "organization",
    "params",
    "type",
    "result_type",
]

REQUIRED_SCORE_FIELDS: list[str] = [
    "overall",
    "spatial",
    "spatio_temporal",
    "temporal",
    "semantic",
]

OPTIONAL_TASK_FIELDS: list[str] = [
    "2d_localization",
    "2d_referring_expressions",
    "2d_spatial_pointing",
    "single_object_tracking",
    "temporal_localization",
    "dense_video_captioning",
    "event_verification",
    "video_qa",
]

ALL_SCORE_FIELDS: list[str] = REQUIRED_SCORE_FIELDS + OPTIONAL_TASK_FIELDS

OPTIONAL_METADATA_FIELDS: list[str] = ["url"]

ALLOWED_TYPES: tuple[str, ...] = ("open", "closed")

ALLOWED_RESULT_TYPES: tuple[str, ...] = ("single", "ensemble")

TYPE_DISPLAY: dict[str, str] = {
    "open": "open",
    "closed": "proprietary",
}


# -- Parameter buckets (legacy list format used by app.py dropdowns) --------

_PARAM_BUCKET_DEFS: list[tuple[str, float | None, float | None]] = [
    ("All",    None, None),
    ("<3B",    0.0,  3.0),
    ("3–10B",  3.0,  10.0),
    ("10–30B", 10.0, 30.0),
    ("30B+",   30.0, None),
]

PARAM_BUCKET_LABELS: list[str] = [b[0] for b in _PARAM_BUCKET_DEFS]


# -- Model type filter -----------------------------------------------------

MODEL_TYPE_LABELS: list[str] = ["All", "Single", "System / Pipeline"]

# UI labels → internal enum value stored in ModelRecord.result_type.
# "System / Pipeline" is the user-facing label; "ensemble" remains the
# internal value for schema/data backwards compatibility.
MODEL_TYPE_TO_INTERNAL: dict[str, str | None] = {
    "All": None,
    "Single": "single",
    "System / Pipeline": "ensemble",
}


# -- Display labels --------------------------------------------------------

PILLAR_LABELS: dict[str, str] = {
    "spatial": "Spatial",
    "spatio_temporal": "Spatio-Temporal",
    "temporal": "Temporal",
    "semantic": "Semantic",
}

PILLAR_LABELS_SHORT: dict[str, str] = {
    "spatial": "Spatial",
    "spatio_temporal": "Sp-Temp",
    "temporal": "Temporal",
    "semantic": "Semantic",
}

TAB_LABELS: dict[str, str] = {
    "overall": "Overall",
    "spatial": "Spatial",
    "spatio_temporal": "Spatio-Temporal",
    "temporal": "Temporal",
    "semantic": "Semantic",
}

TASK_LABELS: dict[str, str] = {
    "2d_localization": "Object Localization",
    "2d_referring_expressions": "Referring Expressions",
    "2d_spatial_pointing": "2D Pointing",
    "single_object_tracking": "Single Object Tracking",
    "temporal_localization": "Temporal Localization",
    "dense_video_captioning": "Dense Video Captioning",
    "event_verification": "Event Verification",
    "video_qa": "Video Question Answering",
}

TASK_LABELS_SHORT: dict[str, str] = {
    "2d_localization":           "Obj.Loc.",
    "2d_referring_expressions":  "Ground",
    "2d_spatial_pointing":       "Point",
    "single_object_tracking":    "SOT",
    "temporal_localization":     "Temp-Loc",
    "dense_video_captioning":    "DVC",
    "event_verification":        "EV",
    "video_qa":                  "VQA",
}

TASK_METRIC_LABELS: dict[str, str] = {
    "2d_localization": "F1@0.5",
    "2d_referring_expressions": "mIoU",
    "2d_spatial_pointing": "Accuracy",
    "single_object_tracking": "AUC",
    "temporal_localization": "mIoU",
    "dense_video_captioning": "SODAc",
    "event_verification": "Macro F1",
    "video_qa": "Accuracy",
}

TASKS_BY_PILLAR: dict[str, list[str]] = {
    "spatial": [
        "2d_localization",
        "2d_referring_expressions",
        "2d_spatial_pointing",
    ],
    "spatio_temporal": [
        "single_object_tracking",
    ],
    "temporal": [
        "temporal_localization",
        "dense_video_captioning",
    ],
    "semantic": [
        "event_verification",
        "video_qa",
    ],
}