Datasets:
File size: 18,636 Bytes
b8ea052 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 | """Compile resolved PVR contracts into executable eval contracts and plans."""
from __future__ import annotations
from collections import Counter, defaultdict
from dataclasses import dataclass
from pathlib import Path
from typing import Any
from pvr_eval.checkers.base import CheckerRegistry
CHECKER_ALIASES = {
"color_blob_parser": "color_classifier",
"grid_slot_parser": "grid_state_parser",
"image_similarity_guard": "input_output_similarity_guard",
"object_zone_checker+preservation_checker": "object_zone_checker|preservation_checker",
}
MODEL_BACKEND_BY_PROPOSAL = {
"dino_box": "dino_or_yolo_detector",
"yolo_box": "yolo_world_or_yoloe_detector",
"sam_mask": "sam_mask_refiner",
"vlm_box": "vlm_region_proposer",
"ocr_layout": "ocr_layout_engine",
"route_mask": "route_mask_parser",
"cv_segmentation": "cv_segmentation_parser",
"grid_parser": "grid_layout_parser",
"color_blob_parser": "color_blob_parser",
}
@dataclass(frozen=True)
class CompileArtifacts:
"""Payloads produced for one resolved case."""
compiled_contract: dict[str, Any]
observation_plan: dict[str, Any]
compile_report: dict[str, Any]
def compile_resolved_case(
*,
source_contract: dict[str, Any],
resolved_contract: dict[str, Any],
source_contract_path: str | Path,
resolved_contract_path: str | Path,
registry: CheckerRegistry | None = None,
) -> CompileArtifacts:
"""Compile one PPAPI-resolved contract into package-native structures.
The resolver emits human-readable ``resolved_checks``. The formal runner
consumes ``atomic_checks``. This bridge preserves the resolver evidence plan
while converting each check into the stable schema used by
:class:`pvr_eval.runners.case_runner.EvaluationRunner`.
"""
case_id = str(resolved_contract.get("case_id") or source_contract.get("case_id") or "").strip()
if not case_id:
raise ValueError("resolved/source contract is missing case_id")
case_dir = Path(source_contract_path).parent
original_questions = _index_original_questions(source_contract)
resolved_checks = [item for item in resolved_contract.get("resolved_checks", []) if isinstance(item, dict)]
atomic_checks = [
_compile_check(item, original_questions.get(str(item.get("id", "")), {}))
for item in resolved_checks
]
compiled_contract = {
"case_id": case_id,
"source": source_contract.get("source", ""),
"top_level": source_contract.get("top_level", ""),
"case_family": resolved_contract.get("case_family") or source_contract.get("case_family", ""),
"task_family": source_contract.get("task_family", ""),
"eval_type": "resolved_executable_contract_v1",
"scoring_policy": source_contract.get("scoring_policy", "task_success_minus_penalty"),
"instruction": source_contract.get("instruction", ""),
"input_image": _resolve_case_path(case_dir, source_contract.get("input_image", "images/input.jpg")),
"gt_image": _resolve_case_path(case_dir, source_contract.get("gt_image", source_contract.get("reference_image", ""))),
"quality_gate": source_contract.get("quality_gate", {}),
"manual_facts": {
**dict(source_contract.get("manual_facts", {}) or {}),
"resolved_intended_final_state": resolved_contract.get("intended_final_state", ""),
},
"zones": source_contract.get("zones", {}),
"metadata": {
"source_contract_path": str(source_contract_path),
"resolved_contract_path": str(resolved_contract_path),
"resolver_version": resolved_contract.get("resolver_version", ""),
"resolver_readiness": resolved_contract.get("readiness", ""),
"global_annotation_needs": resolved_contract.get("global_annotation_needs", []),
"global_auto_region_proposals": resolved_contract.get("global_auto_region_proposals", []),
"audit_notes": resolved_contract.get("audit_notes", []),
},
"atomic_checks": atomic_checks,
}
observation_plan = build_observation_plan(
compiled_contract=compiled_contract,
resolved_contract=resolved_contract,
source_contract_path=source_contract_path,
)
compile_report = build_compile_report(
compiled_contract=compiled_contract,
resolved_contract=resolved_contract,
observation_plan=observation_plan,
registry=registry,
)
return CompileArtifacts(
compiled_contract=compiled_contract,
observation_plan=observation_plan,
compile_report=compile_report,
)
def build_observation_plan(
*,
compiled_contract: dict[str, Any],
resolved_contract: dict[str, Any],
source_contract_path: str | Path,
) -> dict[str, Any]:
"""Build model/parser jobs required before scoring can run."""
case_id = compiled_contract["case_id"]
source_contract_path = Path(source_contract_path)
case_dir = source_contract_path.parent
global_regions = _normalize_region_jobs(
resolved_contract.get("global_auto_region_proposals", []),
source="global_auto_region_proposals",
)
per_check_regions: list[dict[str, Any]] = []
checker_jobs: list[dict[str, Any]] = []
for check in compiled_contract.get("atomic_checks", []):
params = dict(check.get("params", {}) or {})
check_regions = _normalize_region_jobs(
params.get("auto_region_proposals", []),
source=f"check:{check.get('id', '')}",
used_by=[check.get("id", "")],
)
per_check_regions.extend(check_regions)
checker_jobs.append({
"check_id": check.get("id", ""),
"role": check.get("role", "reward"),
"dimension": check.get("dimension", ""),
"checker": check.get("checker", ""),
"original_recommended_checker": params.get("recommended_checker", ""),
"fallback_checker": params.get("fallback_checker", ""),
"region_ids": sorted(_region_ids_for_check(check, check_regions, global_regions)),
"required_evidence": params.get("evidence_required", []),
"decision_rule": params.get("decision_rule", ""),
"params": params.get("checker_params", {}),
})
all_region_jobs = _dedupe_region_jobs(global_regions + per_check_regions)
by_method: dict[str, list[dict[str, Any]]] = defaultdict(list)
for job in all_region_jobs:
by_method[str(job.get("proposal_method", "unknown"))].append(job)
return {
"case_id": case_id,
"readiness": resolved_contract.get("readiness", ""),
"case_family": compiled_contract.get("case_family", ""),
"instruction": compiled_contract.get("instruction", ""),
"case_dir": str(case_dir),
"input_image": _resolve_case_path(case_dir, compiled_contract.get("input_image", "")),
"reference_image": _resolve_case_path(case_dir, compiled_contract.get("gt_image", "")),
"region_jobs": all_region_jobs,
"region_jobs_by_method": dict(sorted(by_method.items())),
"checker_jobs": checker_jobs,
"backend_requirements": _backend_requirements(all_region_jobs, checker_jobs),
"human_annotation_needs": resolved_contract.get("global_annotation_needs", []),
"notes": [
"This plan does not contain model observations yet.",
"Run detector/OCR/SAM/parser backends to materialize observations, then execute compiled_eval_contract.json.",
],
}
def build_compile_report(
*,
compiled_contract: dict[str, Any],
resolved_contract: dict[str, Any],
observation_plan: dict[str, Any],
registry: CheckerRegistry | None = None,
) -> dict[str, Any]:
"""Summarize compile health and execution gaps."""
checker_counts = Counter(str(item.get("checker", "")) for item in compiled_contract.get("atomic_checks", []))
registered: dict[str, bool] = {}
selected: dict[str, str] = {}
if registry is not None:
for checker in checker_counts:
selected_checker = _select_registered_checker(checker, registry)
selected[checker] = selected_checker or ""
registered[checker] = bool(selected_checker)
missing = [checker for checker, ok in registered.items() if not ok]
method_counts = Counter(str(item.get("proposal_method", "unknown")) for item in observation_plan.get("region_jobs", []))
annotation_needs = resolved_contract.get("global_annotation_needs", []) or []
status = "compiled"
if not compiled_contract.get("atomic_checks"):
status = "skipped_no_resolved_checks"
elif missing:
status = "compiled_with_missing_checker"
elif annotation_needs:
status = "compiled_needs_human_annotation"
return {
"case_id": compiled_contract.get("case_id", ""),
"status": status,
"resolver_readiness": resolved_contract.get("readiness", ""),
"atomic_check_count": len(compiled_contract.get("atomic_checks", [])),
"reward_check_count": sum(1 for item in compiled_contract.get("atomic_checks", []) if item.get("role") != "penalty"),
"penalty_check_count": sum(1 for item in compiled_contract.get("atomic_checks", []) if item.get("role") == "penalty"),
"checker_counts": dict(sorted(checker_counts.items())),
"checker_registered": registered,
"checker_selected": selected,
"missing_checkers": missing,
"region_job_count": len(observation_plan.get("region_jobs", [])),
"region_job_method_counts": dict(sorted(method_counts.items())),
"backend_requirements": observation_plan.get("backend_requirements", {}),
"human_annotation_need_count": len(annotation_needs),
"human_annotation_needs": annotation_needs,
"audit_notes": resolved_contract.get("audit_notes", []),
}
def _compile_check(resolved_check: dict[str, Any], original: dict[str, Any]) -> dict[str, Any]:
check_id = str(resolved_check.get("id") or original.get("id") or "").strip()
if not check_id:
raise ValueError("resolved check is missing id")
role = "penalty" if str(resolved_check.get("role") or original.get("role") or "").lower() == "penalty" else "reward"
score_key = "max_deduct" if role == "penalty" else "points"
raw_weight = resolved_check.get("weight_or_max_deduct", original.get(score_key, original.get("weight", 1.0)))
checker = _normalize_checker(str(resolved_check.get("recommended_checker") or original.get("checker") or "observation"))
checker_params = dict(resolved_check.get("params", {}) or {})
return {
"id": check_id,
"role": role,
"dimension": str(resolved_check.get("dimension") or original.get("dimension") or "target_completion"),
score_key: float(raw_weight),
"checker": checker,
"question": original.get("question", ""),
"description": original.get("question") or resolved_check.get("decision_rule", ""),
"pass_criteria": original.get("pass_criteria", ""),
"partial_criteria": original.get("partial_criteria", ""),
"fail_criteria": original.get("fail_criteria", ""),
"hard_fail": bool(original.get("hard_fail", False)),
"params": {
"checker_params": checker_params,
**checker_params,
"original_checker": resolved_check.get("original_checker", original.get("checker", "")),
"recommended_checker": resolved_check.get("recommended_checker", ""),
"normalized_checker": checker,
"fallback_checker": resolved_check.get("fallback_checker", ""),
"region_resolution": resolved_check.get("region_resolution", ""),
"auto_region_proposals": resolved_check.get("auto_region_proposals", []),
"annotation_needs": resolved_check.get("annotation_needs", []),
"needs_annotation": bool(resolved_check.get("needs_annotation", False)),
"decision_rule": resolved_check.get("decision_rule", ""),
"evidence_required": resolved_check.get("evidence_required", []),
"resolver_confidence": resolved_check.get("confidence", None),
},
}
def _index_original_questions(source_contract: dict[str, Any]) -> dict[str, dict[str, Any]]:
out: dict[str, dict[str, Any]] = {}
for role, key in (("reward", "reward_questions"), ("penalty", "penalty_questions")):
for item in source_contract.get(key, []) or []:
if not isinstance(item, dict):
continue
check_id = str(item.get("id", "")).strip()
if check_id:
out[check_id] = {"role": role, **item}
for item in source_contract.get("atomic_checks", []) or []:
if not isinstance(item, dict):
continue
check_id = str(item.get("id") or item.get("check_id") or "").strip()
if check_id and check_id not in out:
out[check_id] = dict(item)
return out
def _normalize_checker(checker: str) -> str:
checker = checker.strip() or "observation"
return CHECKER_ALIASES.get(checker, checker)
def _normalize_region_jobs(
proposals: Any,
*,
source: str,
used_by: list[str] | None = None,
) -> list[dict[str, Any]]:
if not isinstance(proposals, list):
return []
jobs: list[dict[str, Any]] = []
for item in proposals:
if not isinstance(item, dict):
continue
region_id = str(item.get("region_id") or item.get("id") or "").strip()
method = str(item.get("proposal_method") or "unknown").strip()
query = str(item.get("query_or_rule") or item.get("query") or "").strip()
description = str(item.get("description") or "").strip()
job_used_by = item.get("used_by", used_by or [])
if isinstance(job_used_by, str):
job_used_by = [job_used_by]
if not region_id:
region_id = _stable_region_id(method, description, query)
jobs.append({
"region_id": region_id,
"proposal_method": method,
"backend": MODEL_BACKEND_BY_PROPOSAL.get(method, "custom_or_vlm_backend"),
"image_scope": _infer_image_scope(region_id, description, query),
"description": description,
"query_or_rule": query,
"used_by": [str(value) for value in (job_used_by or [])],
"source": source,
})
return jobs
def _dedupe_region_jobs(jobs: list[dict[str, Any]]) -> list[dict[str, Any]]:
merged: dict[tuple[str, str], dict[str, Any]] = {}
for job in jobs:
key = (str(job.get("region_id", "")), str(job.get("proposal_method", "")))
if key not in merged:
merged[key] = dict(job)
continue
existing = merged[key]
existing["used_by"] = sorted(set(existing.get("used_by", [])) | set(job.get("used_by", [])))
if not existing.get("query_or_rule") and job.get("query_or_rule"):
existing["query_or_rule"] = job["query_or_rule"]
if not existing.get("description") and job.get("description"):
existing["description"] = job["description"]
return sorted(merged.values(), key=lambda item: (str(item.get("proposal_method", "")), str(item.get("region_id", ""))))
def _region_ids_for_check(
check: dict[str, Any],
check_regions: list[dict[str, Any]],
global_regions: list[dict[str, Any]],
) -> set[str]:
check_id = str(check.get("id", ""))
out = {str(job.get("region_id", "")) for job in check_regions if job.get("region_id")}
for job in global_regions:
used_by = {str(item) for item in job.get("used_by", []) or []}
if check_id in used_by:
out.add(str(job.get("region_id", "")))
params = check.get("params", {}) or {}
for key, value in params.items():
if key.endswith("_zone") or key.endswith("_region") or key.endswith("_mask") or key in {"slot", "zones", "target_zone"}:
if isinstance(value, str):
out.add(value)
elif isinstance(value, list):
out.update(str(item) for item in value if isinstance(item, str))
return {item for item in out if item}
def _backend_requirements(region_jobs: list[dict[str, Any]], checker_jobs: list[dict[str, Any]]) -> dict[str, Any]:
proposal_counts = Counter(str(item.get("proposal_method", "unknown")) for item in region_jobs)
backend_counts = Counter(str(item.get("backend", "unknown")) for item in region_jobs)
checker_counts = Counter(str(item.get("checker", "unknown")) for item in checker_jobs)
return {
"proposal_method_counts": dict(sorted(proposal_counts.items())),
"backend_counts": dict(sorted(backend_counts.items())),
"checker_counts": dict(sorted(checker_counts.items())),
}
def _select_registered_checker(checker_spec: str, registry: CheckerRegistry) -> str | None:
for candidate in [part.strip() for part in str(checker_spec).split("|") if part.strip()]:
if registry.get(candidate) is not None:
return candidate
base = candidate.split(":", 1)[0].strip()
if registry.get(base) is not None:
return base
return checker_spec if registry.get(checker_spec) is not None else None
def _infer_image_scope(region_id: str, description: str, query: str) -> str:
text = f"{region_id} {description} {query}".lower()
has_input = any(token in text for token in ("input", "source", "original", "before edit"))
has_output = any(token in text for token in ("output", "newly", "drawn", "after edit", "prediction"))
if has_input and has_output:
return "input_and_output"
if has_input:
return "input"
if has_output:
return "output"
return "output_or_input_context"
def _stable_region_id(method: str, description: str, query: str) -> str:
text = "_".join((method, description, query)).lower()
chars = [ch if ch.isalnum() else "_" for ch in text]
token = "_".join("".join(chars).split("_")[:8]).strip("_")
return token or "unnamed_region"
def _resolve_case_path(case_dir: Path, path_text: str) -> str:
if not path_text:
return ""
path = Path(path_text)
if path.is_absolute():
return str(path)
candidate = case_dir / path
if candidate.exists():
return str(candidate.resolve())
image_candidate = case_dir / "images" / path.name
if image_candidate.exists():
return str(image_candidate.resolve())
return str(candidate.resolve())
|