5eb4bdf6c8
The Wikimedia collector truncated the whole basename at 60 characters, so 13 files lost their extension and cv2 could not read them by name - the distractor sets were quietly scoring 13 images fewer than they held. It now truncates the stem and keeps the extension, and the files on disk are renamed to match. The search terms also pull in .djvu volumes whose page 1 Commons serves as a JPEG thumbnail. Twelve book scans were in the sets; they contributed 0 false accepts, and they are not shelves, so the collector filters them and the manifests no longer list them. With all 2650 crops readable the sweep is re-measured end to end: 3.09% at the old lines, 0.30% at the shipped 4.0/4, image6 44 shown / 42 correct / 95.5%. The accept sweep table and the verdict's far-unknown sentence now carry those numbers instead of the pre-calibration ones. Also: the fetched images are gitignored (manifests and results are not), the l10/l11 shelf-tag crops that back the catalog corrections are committed, and a stale pre-gate eval.json and a duplicate of eval/labels/image6.json are gone.
54 lines
2.1 KiB
Python
54 lines
2.1 KiB
Python
"""What the falsely-accepted crops actually read, and what the text lane scores.
|
|
|
|
The advisory says crops 82-97 are Kewpie Mayonnaise 450g - a product the
|
|
catalog does not hold - and that no catalog row can therefore win the text
|
|
ranking, so the veto at DecisionPolicy.decide never fires. This prints the
|
|
reads and the scores that claim depends on.
|
|
"""
|
|
import csv
|
|
import sys
|
|
|
|
import cv2
|
|
import numpy as np
|
|
|
|
sys.path.insert(0, "code")
|
|
from identity_lanes import TextLane, carries_identity # noqa: E402
|
|
from ocr import build_ocr_engine # noqa: E402
|
|
|
|
INDICES = [73, 82, 83, 84, 86, 94, 95, 96, 97]
|
|
|
|
rows = []
|
|
with open("stock-lopia-clean/stock.csv", encoding="utf-8") as handle:
|
|
for record in csv.reader(handle):
|
|
if len(record) >= 2 and record[0].startswith("sku_"):
|
|
rows.append({"sku_id": record[0], "product_name": record[1]})
|
|
|
|
engine = build_ocr_engine("vision")
|
|
lane = TextLane.from_catalog(rows, "stock-lopia-clean", ocr_engine=engine)
|
|
|
|
labels = {f["index"]: f for f in __import__("json").load(open("eval/labels/image6.json"))["facings"]}
|
|
image = cv2.imread("add/image6.jpg")
|
|
|
|
tiles = []
|
|
for index in INDICES:
|
|
facing = labels[index]
|
|
x1, y1, x2, y2 = facing["box"]
|
|
crop = image[y1:y2, x1:x2]
|
|
read = lane.read(crop)
|
|
scores = lane.scores_from_text(read)
|
|
top = sorted(scores.items(), key=lambda kv: -kv[1])[:3]
|
|
print(f"#{index:>3} truth={facing['label']:<9} carries={carries_identity(read)}")
|
|
print(f" read: {read[:110]}")
|
|
print(f" text top: {[(s, round(v, 3)) for s, v in top]}")
|
|
|
|
height, width = crop.shape[:2]
|
|
scale = 260.0 / max(1, height)
|
|
resized = cv2.resize(crop, (max(1, int(width * scale)), 260), interpolation=cv2.INTER_CUBIC)
|
|
canvas = np.full((300, 220, 3), 255, np.uint8)
|
|
canvas[30:290, : min(resized.shape[1], 220)] = resized[:, :220]
|
|
cv2.putText(canvas, str(index), (6, 22), cv2.FONT_HERSHEY_SIMPLEX, 0.7, (0, 0, 255), 2)
|
|
tiles.append(canvas)
|
|
|
|
grid = np.vstack([np.hstack(tiles[:5]), np.hstack(tiles[5:])])
|
|
cv2.imwrite("data/evaluations/2026-09-29-session/scratch/fa_tiles.jpg", grid)
|
|
print("tile:", grid.shape)
|