Datasets:
File size: 12,580 Bytes
4f3323a b6f22ac 4f3323a b6f22ac 4f3323a b6f22ac 4f3323a b6f22ac 4f3323a | 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 | #!/usr/bin/env python3
"""
Anomaly multiplicity per class: how many anomaly types share an image, broken
down by the class present in it. Same visual format as plot_multimodal_aupro.py:
seaborn whitegrid, viridis grouped bars, red dotted class means.
x axis the anchor classes
hue multiplicity 1..7, i.e. the total number of distinct anomaly
types present in the image
"""
import argparse
import sys
from pathlib import Path
import matplotlib
matplotlib.use("Agg")
import matplotlib.pyplot as plt
from matplotlib.patches import Patch
import numpy as np
import pandas as pd
import seaborn as sns
from plot_difficulty import load_annotations, load_scores
DEFAULT_CLASSES = ["residual_content", "nuisance_sticker_inside",
"nuisance_gunk", "nuisance_paper"]
def coco_files(entries, pattern):
found = []
for item in entries:
path = Path(item)
if path.is_dir():
found.extend(path.rglob(pattern))
elif path.is_file():
found.append(path)
return sorted(set(str(path) for path in found))
def type_label(k):
return f"{k} type" if k == 1 else f"{k} types"
def count_table(memberships, classes, max_k):
rows = []
for types in memberships.values():
if not types:
continue
k = len(types)
if k < 1 or k > max_k:
continue
for cls in classes:
if cls in types:
rows.append({"Class": cls, "Multiplicity": k})
if not rows:
return None, None
df = pd.DataFrame(rows)
agg = (df.groupby(["Class", "Multiplicity"]).size()
.reset_index(name="Value"))
agg["n"] = agg["Value"]
means = agg.groupby("Class")["Value"].mean()
return agg, means
def score_table(memberships, scores, classes, max_k):
rows = []
for key, types in memberships.items():
if not types or key not in scores:
continue
k = len(types)
if k < 1 or k > max_k:
continue
for cls in classes:
if cls in types:
rows.append({"Class": cls, "Multiplicity": k,
"Value": scores[key]})
if not rows:
return None, None
df = pd.DataFrame(rows)
agg = (df.groupby(["Class", "Multiplicity"])["Value"]
.agg(["mean", "count"]).reset_index()
.rename(columns={"mean": "Value", "count": "n"}))
means = df.groupby("Class")["Value"].mean()
return agg, means
def add_class_means(classes, class_means, ylim, fmt="{:.2f}"):
for i, cls in enumerate(classes):
if cls not in class_means:
continue
mean_val = float(class_means[cls])
plt.hlines(y=mean_val, xmin=i - 0.4, xmax=i + 0.4,
colors="red", linestyles=":", linewidth=3,
label="Class Mean" if i == 0 else "")
plt.text(i + 0.3, mean_val + 0.018 * (ylim[1] - ylim[0]),
fmt.format(mean_val), color="red", ha="left", va="center",
fontsize=9, fontweight="bold")
def draw(df_agg, class_means, classes, hue_col, hue_order, ylim, ylabel,
stem, annotate_n=False, integer_y=False, mean_fmt="{:.2f}",
legend_labels=None):
sns.set_theme(style="whitegrid", font_scale=1.1)
plt.figure(figsize=(14, 5))
chart = sns.barplot(
data=df_agg,
x="Class",
y="Value",
order=classes,
hue=hue_col,
hue_order=hue_order,
palette="viridis",
)
add_class_means(classes, class_means, ylim, fmt=mean_fmt)
if annotate_n and "n" in df_agg.columns:
for p in chart.patches:
h = p.get_height()
if not np.isfinite(h) or h <= 0:
continue
chart.annotate(f"{int(round(h))}",
(p.get_x() + p.get_width() / 2, h),
ha="center", va="bottom", fontsize=7, color="0.25",
xytext=(0, 2), textcoords="offset points")
plt.xticks(rotation=15, ha="right")
plt.xlabel("")
plt.ylabel(ylabel)
plt.ylim(ylim)
if integer_y:
plt.gca().yaxis.set_major_locator(
matplotlib.ticker.MaxNLocator(integer=True)
)
handles, labels = plt.gca().get_legend_handles_labels()
if legend_labels:
labels = [legend_labels.get(l, l) for l in labels]
plt.legend(
handles, labels,
loc="upper left",
ncol=4,
frameon=True,
framealpha=0.2,
facecolor="white",
edgecolor="black",
)
plt.tight_layout()
plt.savefig(f"{stem}.pdf", bbox_inches="tight")
plt.savefig(f"{stem}.png", dpi=300, bbox_inches="tight")
plt.close()
def report(label, agg, classes, ks, value_name):
print(f"\n{label}")
pivot = (agg.pivot(index="Class", columns="Multiplicity", values="Value")
.reindex(index=classes, columns=ks))
if value_name == "images":
print(" images per group")
print(pivot.fillna(0).astype(int).to_string().replace("\n", "\n "))
print(f" total images (class occurrences): {int(agg['Value'].sum())}")
else:
pivot_n = (agg.pivot(index="Class", columns="Multiplicity", values="n")
.reindex(index=classes, columns=ks))
print(" images per group")
print(pivot_n.fillna(0).astype(int).to_string().replace("\n", "\n "))
print(" mean AUPRO per group")
print(pivot.round(3).to_string().replace("\n", "\n "))
def default_coco_search():
candidates = [
Path("test"),
Path("../test"),
Path(__file__).resolve().parent.parent / "test",
Path("train"),
Path("../train"),
Path(__file__).resolve().parent.parent / "train",
Path("."),
Path(".."),
Path(__file__).resolve().parent.parent,
]
for c in candidates:
if c.is_dir() and list(c.rglob("annotations*.coco.json")):
return [str(c)]
return ["."]
def find_files(names):
resolved = []
for name in names:
candidates = [
Path(f"checkpoint_data/{name}"),
Path(f"../checkpoint_data/{name}"),
Path(__file__).resolve().parent.parent / "checkpoint_data" / name,
Path(__file__).resolve().parent / name,
Path(name),
]
found = next((str(c) for c in candidates if c.is_file()), None)
if found:
resolved.append(found)
return resolved
def default_out_dir():
if Path("checkpoint_data").is_dir():
return Path("checkpoint_data/multiplicity_figures")
if (Path(__file__).resolve().parent.parent / "checkpoint_data").is_dir():
return Path(__file__).resolve().parent.parent / "checkpoint_data/multiplicity_figures"
return Path("./multiplicity_figures")
def main():
parser = argparse.ArgumentParser(description="Anomaly multiplicity per class")
parser.add_argument("--coco", nargs="+", default=None,
help="COCO annotation files or root directory (auto-detected if omitted)")
parser.add_argument("--coco-pattern", default="annotations*.coco.json")
parser.add_argument("--scores", action="append", default=[],
metavar="NAME=CSV[,CSV]",
help="repeatable: per-image AUPRO CSVs for one method. "
"Auto-detected from checkpoint_data/ if omitted.")
parser.add_argument("--metric", default="aupro_30")
parser.add_argument("--classes", nargs="+", default=DEFAULT_CLASSES,
help="anchor classes, in the order they appear on x")
parser.add_argument("--max-k", type=int, default=7,
help="highest multiplicity shown (default 7)")
parser.add_argument("--min-n", type=int, default=1,
help="groups with fewer images are left empty")
parser.add_argument("--annotate-n", action="store_true",
help="print the image count above each bar")
parser.add_argument("--no-shared-y", action="store_true",
help="let each method figure scale its own y axis")
parser.add_argument("--out", default=None,
help="Output directory (default: checkpoint_data/multiplicity_figures or ./multiplicity_figures)")
args = parser.parse_args()
coco_entries = args.coco if args.coco is not None else default_coco_search()
annotations = coco_files(coco_entries, args.coco_pattern)
if not annotations:
raise SystemExit(f"No COCO annotation files found in {coco_entries}")
memberships = load_annotations(annotations)
annotated = {key for key, types in memberships.items() if types}
ks = list(range(1, args.max_k + 1))
out = Path(args.out) if args.out is not None else default_out_dir()
out.mkdir(parents=True, exist_ok=True)
count_agg, count_means = count_table(memberships, args.classes, args.max_k)
if count_agg is None:
raise SystemExit("none of the requested classes occur in the data")
classes = [c for c in args.classes if c in set(count_agg["Class"])]
missing = [c for c in args.classes if c not in classes]
if missing:
print(f"! not present in the data: {', '.join(missing)}", file=sys.stderr)
scores_specs = args.scores
if not scores_specs:
m3dm_csvs = find_files(["per_image_aupro_zivid.csv", "per_image_aupro_realsense.csv"])
stfpm_csvs = find_files(["per_image_aupro_stfpm.csv"])
if m3dm_csvs:
scores_specs.append(f"M3DM={','.join(m3dm_csvs)}")
if stfpm_csvs:
scores_specs.append(f"STFPM={','.join(stfpm_csvs)}")
per_method = {}
for spec in scores_specs:
if "=" not in spec:
raise SystemExit(f"--scores needs NAME=CSV, got {spec!r}")
name, paths = spec.split("=", 1)
files = [p.strip() for p in paths.split(",") if p.strip()]
per_method[name.strip()] = load_scores(files, args.metric, name.strip())
if per_method:
print()
for name, scores in per_method.items():
print(f" {name:<20} {len(annotated & set(scores))} images matched to types")
method_tables = {}
for index, (name, scores) in enumerate(per_method.items(), start=2):
agg, means = score_table(memberships, scores, classes, args.max_k)
if agg is None:
print(f"! {name}: no matching images, skipped", file=sys.stderr)
continue
thin = agg[agg["n"] < args.min_n]
if len(thin):
print(f" {name}: {len(thin)} group(s) below --min-n {args.min_n}, left empty")
agg = agg[agg["n"] >= args.min_n]
slug = name.lower().replace(" ", "_").replace("/", "_")
method_tables[f"{index:02d}_multiplicity_{slug}"] = (name, agg, means)
method_ylim = None
if method_tables and not args.no_shared_y:
top = max(max(float(agg["Value"].max()), float(means.max()))
for _, agg, means in method_tables.values())
method_ylim = (0, top * 1.42)
thin_counts = count_agg[count_agg["n"] < args.min_n]
if len(thin_counts):
count_agg = count_agg[count_agg["n"] >= args.min_n]
report("dataset-wide (composition)", count_agg, classes, ks, "images")
count_top = max(float(count_agg["Value"].max()), float(count_means.max()))
type_names = {k: type_label(k) for k in ks}
draw(count_agg, count_means, classes, "Multiplicity", ks,
(0, count_top * 1.42), "Number of Images",
str(out / "01_multiplicity_dataset_wide"),
annotate_n=False, integer_y=True, mean_fmt="{:.0f}",
legend_labels={str(k): v for k, v in type_names.items()})
count_agg.sort_values(["Class", "Multiplicity"]).to_csv(
out / "01_multiplicity_dataset_wide_values.csv", index=False)
for stem, (label, agg, means) in method_tables.items():
report(label, agg, classes, ks, "aupro")
ylim = method_ylim
if ylim is None:
top = max(float(agg["Value"].max()), float(means.max()))
ylim = (0, top * 1.42)
draw(agg, means, classes, "Multiplicity", ks, ylim,
"Mean Pixel AUPRO Score", str(out / stem),
annotate_n=args.annotate_n,
legend_labels={str(k): v for k, v in type_names.items()})
agg.sort_values(["Class", "Multiplicity"]).to_csv(
out / f"{stem}_values.csv", index=False)
print(f"\nWrote figures and CSVs to {out}")
if __name__ == "__main__":
main()
|