Datasets:

Modalities:
Image
Text
Formats:
parquet
Languages:
Armenian
Size:
< 1K
ArXiv:
License:
ACoPPer / evaluation_kit /evaluation /measure_overall_accuracy.py
Yeva's picture
Add standalone evaluation kit
551cc83 verified
Raw
History Blame Contribute Delete
17.8 kB
#!/usr/bin/env python3
"""Aggregate row-grouping accuracy across matched CSV/JSON page pairs."""
from __future__ import annotations
import argparse
import json
from pathlib import Path
from typing import Any
from measure_accuracy import (
HIGH_IMPACT_REGION_EXAMPLE_COUNT,
build_region_summary,
evaluate_rows,
load_annotation_boxes,
load_predicted_rows,
parse_filter_names,
report_excluded_labels,
safe_mean,
safe_rate,
select_no_box_failure_examples,
summarize_region_example,
)
def parse_args() -> argparse.Namespace:
parser = argparse.ArgumentParser(description=__doc__)
parser.add_argument(
"--data-dir",
type=Path,
help="Directory containing both prediction CSVs and annotation JSONs.",
)
parser.add_argument(
"--predictions-dir",
type=Path,
dest="predictions_dir",
help="Directory containing prediction CSVs (overrides --data-dir for CSVs).",
)
parser.add_argument(
"--annotations-dir",
type=Path,
dest="annotations_dir",
help="Directory containing annotation JSONs (overrides --data-dir for JSONs).",
)
parser.add_argument(
"--coverage-threshold",
type=float,
default=1.0,
help="Minimum fraction of words in a row that must fit a box for a full match.",
)
parser.add_argument(
"--failure-example-count",
type=int,
default=5,
help="Number of aggregate failure examples to keep per failure type.",
)
parser.add_argument(
"--filter",
dest="filters",
help=(
"Optional comma-separated region filters, e.g. "
"`non-armenian`, `graphics`, or `non-armenian,graphics`."
),
)
parser.add_argument(
"--unit-level",
dest="unit_level",
choices=["word", "line"],
default="word",
help="Granularity of predicted rows: 'word' or 'line'.",
)
parser.add_argument(
"--output",
type=Path,
help="Optional JSON path for the aggregated report.",
)
return parser.parse_args()
def discover_pairs(
predictions_dir: Path,
annotations_dir: Path,
*,
partial: bool = False,
) -> list[tuple[str, Path, Path]]:
csv_by_stem = {path.stem: path for path in sorted(predictions_dir.glob("*.csv"))}
json_by_stem = {
"_".join(path.relative_to(annotations_dir).with_suffix("").parts): path
for path in sorted(annotations_dir.rglob("*.json"))
if not path.stem.endswith("_result")
}
missing_csv = sorted(json_by_stem.keys() - csv_by_stem.keys())
missing_json = sorted(csv_by_stem.keys() - json_by_stem.keys())
problems: list[str] = []
if missing_csv:
if partial:
print(
f"--partial: skipping {len(missing_csv)} page(s) with no prediction CSV",
flush=True,
)
else:
problems.append(f"missing CSV for: {', '.join(missing_csv)}")
if missing_json:
problems.append(f"missing JSON for: {', '.join(missing_json)}")
if problems:
raise SystemExit(
f"Unmatched files (CSVs in {predictions_dir}, JSONs in {annotations_dir}): "
f"{'; '.join(problems)}"
)
pair_names = sorted(csv_by_stem.keys() & json_by_stem.keys())
if not pair_names:
raise SystemExit(
f"No matched CSV/JSON pairs found "
f"(CSVs in {predictions_dir}, JSONs in {annotations_dir})"
)
return [(name, csv_by_stem[name], json_by_stem[name]) for name in pair_names]
def enrich_record(record: dict[str, Any], *, page_name: str, predictions_csv: Path, annotations_json: Path) -> dict[str, Any]:
enriched = dict(record)
enriched["page_name"] = page_name
enriched["predictions_csv"] = str(predictions_csv)
enriched["annotations_json"] = str(annotations_json)
return enriched
def empty_filter_accumulator(page_filter: dict[str, Any]) -> dict[str, Any]:
return {
"filters": page_filter.get("filters", []),
"text_box_count": 0,
"excluded_box_count": 0,
"included_box_count": 0,
"threshold": page_filter.get("threshold"),
"labels": page_filter.get("labels"),
}
def update_filter_accumulator(
accumulator: dict[str, Any],
page_filter: dict[str, Any],
) -> None:
accumulator["text_box_count"] += page_filter.get("text_box_count", 0)
accumulator["excluded_box_count"] += page_filter.get("excluded_box_count", 0)
accumulator["included_box_count"] += page_filter.get("included_box_count", 0)
if accumulator["threshold"] is None and "threshold" in page_filter:
accumulator["threshold"] = page_filter["threshold"]
if accumulator["labels"] is None and "labels" in page_filter:
accumulator["labels"] = page_filter["labels"]
def summarize_filter_accumulator(accumulator: dict[str, Any]) -> dict[str, Any]:
summary = {
"filters": accumulator["filters"],
"text_box_count": accumulator["text_box_count"],
"excluded_box_count": accumulator["excluded_box_count"],
"excluded_box_rate": safe_rate(
accumulator["excluded_box_count"],
accumulator["text_box_count"],
),
"included_box_count": accumulator["included_box_count"],
}
if accumulator["threshold"] is not None:
summary["threshold"] = accumulator["threshold"]
if accumulator["labels"] is not None:
summary["labels"] = accumulator["labels"]
return summary
def build_aggregate_failure_examples(
rows: list[dict[str, Any]],
split_line_groups: list[dict[str, Any]],
ocr_regions: list[dict[str, Any]],
example_count: int,
) -> dict[str, list[dict[str, Any]]]:
def simplify(row: dict[str, Any]) -> dict[str, Any]:
simplified = {
"page_name": row["page_name"],
"predictions_csv": row["predictions_csv"],
"annotations_json": row["annotations_json"],
"row_id": row["row_id"],
"row_text": row["row_text"],
"dominant_box_id": row["dominant_box_id"],
"dominant_coverage": row["dominant_coverage"],
"touched_box_ids": row["touched_box_ids"],
"candidate_boxes": row["per_box_coverages"][:3],
}
if row["status"] == "no_box":
simplified["single_uncovered_word_against_dominant_box"] = row[
"single_uncovered_word_against_dominant_box"
]
simplified["uncovered_words_against_dominant_box"] = row[
"uncovered_words_against_dominant_box"
]
return simplified
multiple_rows = [row for row in rows if row["status"] == "multiple_boxes"]
no_box_rows = [row for row in rows if row["status"] == "no_box"]
detected_empty_rows = [row for row in rows if row.get("is_detected_empty")]
multiple_examples = sorted(
multiple_rows,
key=lambda row: (
-len(row["touched_box_ids"]),
row["dominant_coverage"],
row["page_name"],
row["row_id"],
),
)[:example_count]
no_box_examples = select_no_box_failure_examples(no_box_rows, example_count)
split_line_examples = sorted(
split_line_groups,
key=lambda group: (group["page_name"], group["box_id"], group["row_ids"][0]),
)
high_impact_examples = sorted(
[
region
for region in ocr_regions
if region["text_metrics"]["char_edit_distance"] > 0
],
key=lambda region: (
-region["text_metrics"]["char_edit_distance"],
region["page_name"],
region["region_id"],
),
)[:HIGH_IMPACT_REGION_EXAMPLE_COUNT]
normal_single_box_error_examples = sorted(
[
region
for region in ocr_regions
if (
region.get("normal_single_box_region")
and region["text_metrics"]["char_edit_distance"] > 0
)
],
key=lambda region: (
-region["text_metrics"]["cer"],
-region["text_metrics"]["char_edit_distance"],
region["page_name"],
region["region_id"],
),
)[:example_count]
return {
"multiple_boxes": [simplify(row) for row in multiple_examples],
"no_box": [simplify(row) for row in no_box_examples],
"detected_empty": [simplify(row) for row in detected_empty_rows[:example_count]],
"split_line": split_line_examples,
"high_impact_regions": [
summarize_region_example(region, include_error_stats=True)
for region in high_impact_examples
],
"normal_single_box_region_errors": [
summarize_region_example(region, include_error_stats=True)
for region in normal_single_box_error_examples
],
}
def aggregate_reports(
page_reports: list[dict[str, Any]],
coverage_threshold: float,
failure_example_count: int,
unit_level: str = "word",
) -> dict[str, Any]:
all_rows: list[dict[str, Any]] = []
all_ignored_rows: list[dict[str, Any]] = []
all_split_line_groups: list[dict[str, Any]] = []
all_filtered_text_boxes: list[dict[str, Any]] = []
all_ocr_regions: list[dict[str, Any]] = []
best_coverages: list[float] = []
total_rows = 0
ignored_watermark_rows = 0
exactly_one_box = 0
multiple_boxes = 0
no_box = 0
detected_empty = 0
split_line = 0
split_line_rows = 0
missing_text_boxes = 0
total_detected_word_boxes = 0
total_gt_box_count = 0
total_gt_char_count = 0
filter_accumulator: dict[str, Any] | None = None
pages: list[dict[str, Any]] = []
for page in page_reports:
name = page["page_name"]
predictions_csv = page["predictions_csv"]
annotations_json = page["annotations_json"]
report = page["report"]
summary = report["summary"]
total_rows += summary["total_rows"]
ignored_watermark_rows += summary["ignored_watermark_rows"]
exactly_one_box += summary["exactly_one_box"]
multiple_boxes += summary["multiple_boxes"]
no_box += summary["no_box"]
detected_empty += summary.get("detected_empty", 0)
split_line += summary.get("split_line", 0)
split_line_rows += summary.get("split_line_rows", 0)
missing_text_boxes += summary.get("missing_text_boxes", 0)
total_detected_word_boxes += summary.get("total_detected_word_boxes", 0)
total_gt_box_count += summary.get("gt_box_count", 0)
total_gt_char_count += summary.get("gt_char_count", 0)
page_filter = summary.get("filter")
if page_filter is not None:
if filter_accumulator is None:
filter_accumulator = empty_filter_accumulator(page_filter)
update_filter_accumulator(filter_accumulator, page_filter)
ignored_rows = [
enrich_record(
row,
page_name=name,
predictions_csv=predictions_csv,
annotations_json=annotations_json,
)
for row in report["ignored_rows"]
]
split_line_groups = [
enrich_record(
group,
page_name=name,
predictions_csv=predictions_csv,
annotations_json=annotations_json,
)
for group in report.get("split_line_groups", [])
]
filtered_text_boxes = [
enrich_record(
box,
page_name=name,
predictions_csv=predictions_csv,
annotations_json=annotations_json,
)
for box in report.get("filtered_text_boxes", [])
]
all_rows.extend(
enrich_record(row, page_name=name, predictions_csv=predictions_csv, annotations_json=annotations_json)
for row in report["rows"]
)
all_ignored_rows.extend(ignored_rows)
all_split_line_groups.extend(split_line_groups)
all_filtered_text_boxes.extend(filtered_text_boxes)
best_coverages.extend(
row["dominant_coverage"]
for row in report["rows"]
if row["status"] != "no_box"
)
all_ocr_regions.extend(
enrich_record(
region,
page_name=name,
predictions_csv=predictions_csv,
annotations_json=annotations_json,
)
for region in report.get("ocr_regions", [])
)
page_entry = {
"page_name": name,
"predictions_csv": str(predictions_csv),
"annotations_json": str(annotations_json),
"summary": summary,
"ocr_regions": report.get("ocr_regions", []),
"split_line_groups": split_line_groups,
"ignored_rows": ignored_rows,
}
if page_filter is not None:
page_entry["filter"] = page_filter
page_entry["filtered_text_boxes"] = filtered_text_boxes
pages.append(page_entry)
region_summary = build_region_summary(all_ocr_regions)
summary = {
"unit_level": unit_level,
"coverage_threshold": coverage_threshold,
"pair_count": len(page_reports),
"total_rows": total_rows,
"ignored_watermark_rows": ignored_watermark_rows,
"exactly_one_box": exactly_one_box,
"exactly_one_box_rate": safe_rate(exactly_one_box, total_rows),
"multiple_boxes": multiple_boxes,
"multiple_boxes_rate": safe_rate(multiple_boxes, total_rows),
"no_box": no_box,
"no_box_rate": safe_rate(no_box, total_rows),
"detected_empty": detected_empty,
"split_line": split_line,
"split_line_rate": safe_rate(split_line, total_rows),
"split_line_rows": split_line_rows,
"split_line_rows_rate": safe_rate(split_line_rows, total_rows),
"mean_best_coverage": safe_mean(best_coverages),
"gt_box_count": total_gt_box_count,
"gt_char_count": total_gt_char_count,
"ocr_region_count": region_summary["ocr_region_count"],
"multibox_region_count": region_summary["multibox_region_count"],
"ocr_region_mean_cer": region_summary["mean_cer"],
"ocr_region_gt_char_count": region_summary["gt_char_count"],
"ocr_region_char_edit_distance": region_summary["char_edit_distance"],
"ocr_region_cer": region_summary["cer"],
"ocr_region_cer_lowercase": region_summary["cer_lowercase"],
"ocr_region_cer_buckets": region_summary["cer_buckets"],
"normal_single_box_region": region_summary["normal_single_box_region"],
"missing_text_boxes": missing_text_boxes,
"total_detected_word_boxes": total_detected_word_boxes,
"missing_text_box_rate": safe_rate(
missing_text_boxes,
total_detected_word_boxes,
),
}
if filter_accumulator is not None:
summary["filter"] = summarize_filter_accumulator(filter_accumulator)
failure_examples = build_aggregate_failure_examples(
rows=all_rows,
split_line_groups=all_split_line_groups,
ocr_regions=all_ocr_regions,
example_count=failure_example_count,
)
aggregate_report = {
"summary": summary,
"pages": pages,
"failure_examples": failure_examples,
"split_line_groups": all_split_line_groups,
"ignored_rows": all_ignored_rows,
"excluded_labels": report_excluded_labels(),
}
if filter_accumulator is not None:
aggregate_report["filtered_text_boxes"] = all_filtered_text_boxes
return aggregate_report
def main() -> None:
args = parse_args()
predictions_dir = (args.predictions_dir or args.data_dir)
annotations_dir = (args.annotations_dir or args.data_dir)
if not predictions_dir or not annotations_dir:
raise SystemExit(
"Provide --data-dir or both --predictions-dir and --annotations-dir."
)
pairs = discover_pairs(predictions_dir.resolve(), annotations_dir.resolve())
try:
filters = parse_filter_names(args.filters)
except ValueError as error:
raise SystemExit(str(error)) from error
page_reports: list[dict[str, Any]] = []
for name, predictions_csv, annotations_json in pairs:
predicted_rows = load_predicted_rows(predictions_csv, unit_level=args.unit_level)
annotation_boxes = load_annotation_boxes(annotations_json)
report = evaluate_rows(
predicted_rows=predicted_rows,
annotation_boxes=annotation_boxes,
coverage_threshold=args.coverage_threshold,
failure_example_count=args.failure_example_count,
hide_zero_cer_details=False,
filters=filters,
unit_level=args.unit_level,
)
page_reports.append(
{
"page_name": name,
"predictions_csv": predictions_csv,
"annotations_json": annotations_json,
"report": report,
}
)
aggregate_report = aggregate_reports(
page_reports=page_reports,
coverage_threshold=args.coverage_threshold,
failure_example_count=args.failure_example_count,
unit_level=args.unit_level,
)
print(json.dumps(aggregate_report["summary"], ensure_ascii=False, indent=2))
if args.output:
args.output.write_text(
json.dumps(aggregate_report, ensure_ascii=False, indent=2),
encoding="utf-8",
)
if __name__ == "__main__":
main()