diff --git a/recognization/code/app.py b/recognization/code/app.py
index fb84113..4b267db 100644
--- a/recognization/code/app.py
+++ b/recognization/code/app.py
@@ -392,6 +392,7 @@ def create_app(
result["compliance"] = bundle["compliance"]
result["ocr_engine"] = ocr_engine.name
result["tag_timing_ms"] = round(elapsed_ms, 1)
+ engine.image_store.set_tag_bundle(result["image_id"], bundle)
return result
@app.post("/v1/images/{image_id}/classify", dependencies=[Depends(authorize)])
@@ -413,22 +414,44 @@ def create_app(
if row is not None:
detection["identity"] = row["identity"]
engine.image_store.update_detections(image_id, stored.detections)
- identities = [
- {
+
+ # Tags were read before the identities existed; reconcile them now.
+ compliance = None
+ if stored.tag_bundle:
+ from reconcile import reconcile_bundle
+
+ catalog_by_sku = {product["sku_id"]: product for product in engine.catalog()}
+ refreshed = reconcile_bundle(stored.tag_bundle, stored.detections, catalog_by_sku)
+ for detection in stored.detections:
+ decision = refreshed["reconciliations"].get(int(detection["id"]))
+ if decision is not None:
+ detection["reconciliation"] = decision
+ compliance = refreshed["compliance"]
+ identities = []
+ for detection in detections:
+ row = {
"detection_id": int(detection["id"]),
"box": list(detection["box"]),
"identity": detection["identity"],
}
- for detection in detections
- ]
+ # Re-reconciled after classification; without it a client would keep
+ # the `tag_only` verdict it got when identities did not exist yet.
+ if detection.get("reconciliation") is not None:
+ row["reconciliation"] = detection["reconciliation"]
+ if detection.get("price_tag") is not None:
+ row["price_tag"] = detection["price_tag"]
+ identities.append(row)
summary = engine.summarize(stored.detections, [])
- return {
+ payload = {
"image_id": image_id,
"method": method,
"identities": identities,
"summary": summary,
"timing_ms": {"classify": round((time.perf_counter() - started) * 1000.0, 2)},
}
+ if compliance is not None:
+ payload["compliance"] = compliance
+ return payload
@app.post("/v1/images/{image_id}/inspect", dependencies=[Depends(authorize)])
async def inspect_image(image_id: str, body: InspectBody) -> Dict[str, Any]:
diff --git a/recognization/code/engine.py b/recognization/code/engine.py
index 273fbf4..92477c2 100644
--- a/recognization/code/engine.py
+++ b/recognization/code/engine.py
@@ -98,6 +98,9 @@ class StoredImage:
detections: List[Dict[str, Any]] = field(default_factory=list)
created_at: float = 0.0
nbytes: int = 0
+ # The tag pass for this still (tags, bands, box->tag), kept so a later
+ # classify can re-reconcile against the identities it just produced.
+ tag_bundle: Optional[Dict[str, Any]] = None
def identity_state_from_level(level: Optional[str]) -> str:
@@ -192,6 +195,17 @@ class ImageStore:
if stored is not None:
stored.detections = list(detections)
+ def set_tag_bundle(self, image_id: str, bundle: Dict[str, Any]) -> None:
+ with self._lock:
+ stored = self._items.get(image_id)
+ if stored is not None:
+ stored.tag_bundle = bundle
+
+ def get_tag_bundle(self, image_id: str) -> Optional[Dict[str, Any]]:
+ with self._lock:
+ stored = self._items.get(image_id)
+ return stored.tag_bundle if stored is not None else None
+
def _evict_locked(self) -> None:
now = time.time()
for image_id in [k for k, v in self._items.items() if now - v.created_at > self.ttl_s]:
diff --git a/recognization/code/reconcile.py b/recognization/code/reconcile.py
index 809a328..00376f3 100644
--- a/recognization/code/reconcile.py
+++ b/recognization/code/reconcile.py
@@ -189,7 +189,9 @@ class ComplianceSettings:
# A row may only be judged when enough of its facings carry tag evidence
# (readable or unreadable). Below this, the row is "not evaluated": claiming a
# tag is missing because our detector found none would be a false violation.
- min_row_tag_coverage: float = 0.5
+ # 0.7 measured: at 0.5 a row where our own tag detection was patchy produced a
+ # false "group without price tag" on a shelf where every facing has a tag.
+ min_row_tag_coverage: float = 0.7
@dataclass
@@ -434,12 +436,23 @@ def attach_tags_to_boxes(
loop to protect - then attaches tags, reconciles, and evaluates compliance.
"""
image_shape = image.shape
- bands = tag_service.plan(boxes, image_shape)
- for band in bands:
- tag_service.run_pass(image, band)
- # Plan again: a still has no later frame to pick up a second band.
- for band in tag_service.plan(boxes, image_shape):
- tag_service.run_pass(image, band)
+ # A still has no later frame to pick up the remaining bands, and `plan` only
+ # hands out a few per call (a stream must stay responsive). Drain them, with a
+ # cap so a huge shelf cannot turn one request into minutes of OCR. Draining
+ # also feeds the vote: a band read in two rounds is two observations.
+ bands: List[Any] = []
+ rounds = 0
+ while rounds < 4:
+ planned = tag_service.plan(boxes, image_shape)
+ if not planned:
+ break
+ for band in planned:
+ # Two independent reads (different upscale) so the vote has evidence
+ # to work with; a single read has nothing to vote against.
+ tag_service.run_pass(image, band, scale_bias=1.0)
+ tag_service.run_pass(image, band, scale_bias=1.7)
+ bands.append(band)
+ rounds += 1
tracks = [
{
@@ -468,6 +481,61 @@ def attach_tags_to_boxes(
"counts": counts,
"reconciliations": reconciliations,
"bands": [{"row_index": band.row_index, "box": band.box} for band in bands],
+ "bands_objects": list(bands),
+ "compliance": {
+ "violations": summary["violations"],
+ "by_rule": summary["by_rule"],
+ "rows": [
+ {
+ "row_index": finding.row_index,
+ "rule": finding.rule,
+ "box": finding.box,
+ "missing": finding.count,
+ "skus": finding.skus,
+ }
+ for finding in findings
+ if finding.severity == "violation"
+ ],
+ "hints": summary["hints"],
+ },
+ }
+
+
+def reconcile_bundle(
+ bundle: Dict[str, Any],
+ detections: Sequence[Dict[str, Any]],
+ catalog_by_sku: Dict[str, Dict[str, Any]],
+) -> Dict[str, Any]:
+ """Re-run reconciliation and compliance against a cached tag bundle.
+
+ Used when identities arrive after the tag pass (photo mode: recognize paints
+ boxes, classify names them). Without this, every tag would look like `tag_only`
+ because it was compared against an identity that did not exist yet.
+ """
+ tracks = [
+ {
+ "track_id": int(detection["id"]),
+ "box": list(detection["box"]),
+ "identity": detection.get("identity") or {},
+ }
+ for detection in detections
+ ]
+ tag_by_track = bundle.get("by_box") or {}
+ reconciliations: Dict[int, Dict[str, Any]] = {}
+ for track in tracks:
+ track_id = track["track_id"]
+ tag = tag_by_track.get(track_id)
+ product = catalog_by_sku.get((track["identity"] or {}).get("sku_id") or "")
+ decision = reconcile(track["identity"], tag, product)
+ if decision is not None:
+ reconciliations[track_id] = decision
+
+ findings = evaluate_compliance(
+ tracks, tag_by_track, bundle.get("tags") or [], bundle.get("bands_objects") or [], reconciliations
+ )
+ summary = summarize_findings(findings)
+ return {
+ "reconciliations": reconciliations,
"compliance": {
"violations": summary["violations"],
"by_rule": summary["by_rule"],
diff --git a/recognization/code/tags.py b/recognization/code/tags.py
index 0bcd93d..902f36c 100644
--- a/recognization/code/tags.py
+++ b/recognization/code/tags.py
@@ -309,7 +309,9 @@ def parse_price_tag(texts: Sequence[str], lines: Optional[Sequence[OcrLine]] = N
value = int(match.group(1))
except ValueError:
continue
- if not (1 <= value <= 99999):
+ # A Japanese shelf tag is ¥10-¥9,999. Five-digit values are misreads
+ # (a packaging size, a unit price, a barcode fragment).
+ if not (10 <= value <= 9999):
continue
area = float(_text_area(line)) if line is not None else float(len(match.group(1)))
price_candidates.append((value, area))
@@ -427,8 +429,13 @@ class TagService:
self._pending_bands.add(key)
return chosen
- def run_pass(self, image: Any, band: RailBand) -> List[TagObservation]:
- """OCR one band and fold the observations into the cache (worker thread)."""
+ def run_pass(self, image: Any, band: RailBand, scale_bias: float = 1.0) -> List[TagObservation]:
+ """OCR one band and fold the observations into the cache (worker thread).
+
+ `scale_bias` lets a caller read the same band at a different upscale: two
+ independent reads of the same tag are what makes the vote meaningful
+ (identical pixels would produce identical output).
+ """
key = (band.row_index, band.box[0] // 32, band.box[2] // 32)
try:
x1, y1, x2, y2 = band.box
@@ -440,6 +447,8 @@ class TagService:
digit_estimate = crop.shape[0]
if digit_estimate < 90:
scale = min(2.5, 120.0 / max(1, digit_estimate))
+ scale = max(1.0, min(3.0, scale * max(0.5, scale_bias)))
+ if scale > 1.0:
import cv2
crop = cv2.resize(crop, None, fx=scale, fy=scale, interpolation=cv2.INTER_CUBIC)
@@ -525,32 +534,40 @@ class TagService:
def _vote(self, record: TagRecord) -> None:
"""Majority vote over observations: the accuracy lever for digit misreads."""
- prices: Dict[int, int] = {}
- regulars: Dict[int, int] = {}
- names: Dict[str, int] = {}
- units: Dict[float, int] = {}
- discounts: Dict[float, int] = {}
+ # Votes are weighted by the read's confidence: two equally common values
+ # are then decided by which read was surer, never by which number is
+ # bigger. A tie with no confident read stays unread.
+ prices: Dict[int, float] = {}
+ regulars: Dict[int, float] = {}
+ names: Dict[str, float] = {}
+ units: Dict[float, float] = {}
+ discounts: Dict[float, float] = {}
flags: set = set()
confidences: List[float] = []
for observation in record.observations:
fields = observation.fields
+ weight = max(0.05, float(observation.confidence))
if fields.price is not None:
- prices[fields.price] = prices.get(fields.price, 0) + 1
+ prices[fields.price] = prices.get(fields.price, 0.0) + weight
if fields.regular_price is not None:
- regulars[fields.regular_price] = regulars.get(fields.regular_price, 0) + 1
+ regulars[fields.regular_price] = regulars.get(fields.regular_price, 0.0) + weight
if fields.name:
- names[fields.name] = names.get(fields.name, 0) + 1
+ names[fields.name] = names.get(fields.name, 0.0) + weight
if fields.unit_price is not None:
- units[fields.unit_price] = units.get(fields.unit_price, 0) + 1
+ units[fields.unit_price] = units.get(fields.unit_price, 0.0) + weight
if fields.discount is not None:
- discounts[fields.discount] = discounts.get(fields.discount, 0) + 1
+ discounts[fields.discount] = discounts.get(fields.discount, 0.0) + weight
flags.update(fields.flags)
confidences.append(observation.confidence)
- def top(counter: Dict[Any, int]) -> Any:
+ def top(counter: Dict[Any, float]) -> Any:
if not counter:
return None
- return max(counter.items(), key=lambda item: (item[1], item[0]))[0]
+ ranked = sorted(counter.items(), key=lambda item: item[1], reverse=True)
+ if len(ranked) > 1 and abs(ranked[0][1] - ranked[1][1]) < 1e-9:
+ # A dead heat between two different readings is not evidence.
+ return None
+ return ranked[0][0]
record.price = top(prices)
record.regular_price = top(regulars)
diff --git a/recognization/code/test_reconcile.py b/recognization/code/test_reconcile.py
index 94d98b2..465628d 100644
--- a/recognization/code/test_reconcile.py
+++ b/recognization/code/test_reconcile.py
@@ -152,15 +152,15 @@ BANDS = [band(1, [0, 300, 700, 340]), band(2, [0, 600, 700, 640])]
class ComplianceTests(unittest.TestCase):
def test_missing_tag_on_a_covered_row_is_a_violation(self):
- tracks = [
- track(1, [10, 100, 90, 300]),
- track(2, [110, 100, 190, 300]),
- ]
+ # 3 of 4 facings carry tag evidence (75% >= the 70% bar), one has none.
+ tracks = [track(index, [10 + index * 150, 100, 140 + index * 150, 300]) for index in range(4)]
tags = {
- 1: tag(state=TAG_STATE_READABLE, price=168),
+ 0: tag(state=TAG_STATE_READABLE, price=168, id="tag_a"),
+ 1: tag(state=TAG_STATE_READABLE, price=178, id="tag_b"),
2: tag(state=TAG_STATE_ABSENT),
+ 3: tag(state=TAG_STATE_READABLE, price=188, id="tag_c"),
}
- findings = evaluate_compliance(tracks, tags, [tags[1]], BANDS)
+ findings = evaluate_compliance(tracks, tags, [tags[0], tags[1], tags[3]], BANDS)
rules = [finding.rule for finding in findings]
self.assertIn(RULE_MISSING_TAG, rules)
missing = [finding for finding in findings if finding.rule == RULE_MISSING_TAG][0]
@@ -201,18 +201,13 @@ class ComplianceTests(unittest.TestCase):
self.assertEqual(findings, [])
def test_group_of_consecutive_facings_without_tags(self):
- # 5 facings: 2 tagged, 3 not. Coverage counts facings with *any* tag
- # evidence, so give the row enough evidence to be judged at all.
- tracks = [track(index, [10 + index * 100, 100, 90 + index * 100, 300]) for index in range(6)]
- tags = {
- 0: tag(state=TAG_STATE_READABLE, price=168, id="tag_a"),
- 1: tag(state=TAG_STATE_READABLE, price=178, id="tag_b"),
- 2: tag(state=TAG_STATE_ABSENT),
- 3: tag(state=TAG_STATE_ABSENT),
- 4: tag(state=TAG_STATE_ABSENT),
- 5: tag(state=TAG_STATE_READABLE, price=188, id="tag_c"),
- }
- findings = evaluate_compliance(tracks, tags, [tags[0], tags[1], tags[5]], BANDS)
+ # Coverage counts facings with *any* tag evidence, so the row needs enough
+ # of it to be judged at all (bar is 70%).
+ tracks = [track(index, [10 + index * 70, 100, 70 + index * 70, 300]) for index in range(10)]
+ tags = {index: tag(state=TAG_STATE_READABLE, price=168 + index, id=f"tag_{index}") for index in range(10)}
+ for absent in (2, 3, 4):
+ tags[absent] = tag(state=TAG_STATE_ABSENT, id=None)
+ findings = evaluate_compliance(tracks, tags, [t for t in tags.values() if t["id"]], BANDS)
groups = [finding for finding in findings if finding.rule == RULE_MISSING_TAG_GROUP]
self.assertEqual(len(groups), 1)
self.assertEqual(groups[0].count, 3)
@@ -220,13 +215,14 @@ class ComplianceTests(unittest.TestCase):
self.assertIn("3 facings", groups[0].detail)
def test_two_missing_facings_are_not_a_group(self):
- tracks = [track(index, [10 + index * 100, 100, 90 + index * 100, 300]) for index in range(3)]
+ tracks = [track(index, [10 + index * 150, 100, 140 + index * 150, 300]) for index in range(4)]
tags = {
0: tag(state=TAG_STATE_READABLE, price=168, id="tag_a"),
1: tag(state=TAG_STATE_ABSENT),
2: tag(state=TAG_STATE_READABLE, price=188, id="tag_b"),
+ 3: tag(state=TAG_STATE_READABLE, price=198, id="tag_c"),
}
- findings = evaluate_compliance(tracks, tags, [tags[0], tags[2]], BANDS)
+ findings = evaluate_compliance(tracks, tags, [tags[0], tags[2], tags[3]], BANDS)
self.assertNotIn(RULE_MISSING_TAG_GROUP, [finding.rule for finding in findings])
missing = [finding for finding in findings if finding.rule == RULE_MISSING_TAG]
self.assertEqual(len(missing), 1)
diff --git a/recognization/code/web/app.js b/recognization/code/web/app.js
index 7dd3be4..6661b0e 100644
--- a/recognization/code/web/app.js
+++ b/recognization/code/web/app.js
@@ -33,6 +33,15 @@ const I18N = {
connectionLost: 'Connection lost — retrying…', unknownMethod: 'Unknown algorithm',
noServer: 'Server unreachable', confirmEmpty: 'Hide empty spaces', showEmpty: 'Show empty spaces',
facings: 'facings', inView: 'in view', products: 'products',
+ ocrTitle: 'Price tag', ocrPrice: 'Tag price', ocrRegular: 'Regular price',
+ ocrDiscount: 'Discount', ocrUnit: 'Unit price', ocrName: 'Tag name',
+ ocrState: 'Tag', ocrUnreadable: 'unreadable — move closer', ocrAbsent: 'no tag under this product',
+ ocrNote: 'Note', row: 'Row', more: 'more',
+ recon_confirmed: 'tag matches', recon_mismatch: 'tag differs', recon_unverified: 'tag unread',
+ recon_no_tag: 'no tag', recon_tag_only: 'tag only',
+ rule_missing_price_tag: 'missing price tag', rule_missing_tag_group: 'group without price tag',
+ rule_orphan_tag: 'tag with no product', rule_price_mismatch: 'price differs from catalog',
+ rule_duplicate_tag: 'tag shared by two products',
},
ja: {
subtitle: 'スマートスーパーマーケット棚認識 & 在庫管理AI',
@@ -56,6 +65,15 @@ const I18N = {
connectionLost: '接続が切れました — 再接続中…', unknownMethod: '不明なアルゴリズム',
noServer: 'サーバーに接続できません', confirmEmpty: '空きスペースを隠す', showEmpty: '空きスペースを表示',
facings: 'フェイス', inView: '表示中', products: '商品',
+ ocrTitle: '値札', ocrPrice: '値札の価格', ocrRegular: '通常価格',
+ ocrDiscount: '割引', ocrUnit: '単位価格', ocrName: '値札の商品名',
+ ocrState: '値札', ocrUnreadable: '読み取れません — 近づいてください', ocrAbsent: '商品の下に値札がありません',
+ ocrNote: '備考', row: '段', more: '件',
+ recon_confirmed: '一致', recon_mismatch: '不一致', recon_unverified: '未読取',
+ recon_no_tag: '値札なし', recon_tag_only: '値札のみ',
+ rule_missing_price_tag: '値札なし', rule_missing_tag_group: '値札のない商品が連続',
+ rule_orphan_tag: '商品のない値札', rule_price_mismatch: 'カタログ価格と不一致',
+ rule_duplicate_tag: '同じ値札が2商品',
},
};
@@ -151,8 +169,13 @@ function render() {
ctx.clearRect(0, 0, overlay.width, overlay.height);
const now = performance.now();
- if (state.mode === 'photo' && state.photoDetections.length) {
- drawPhotoDetections();
+ if (state.mode === 'photo') {
+ // Draw the photo into the canvas so annotations cannot drift from it.
+ const photo = el('photo');
+ if (photo.complete && photo.naturalWidth) {
+ ctx.drawImage(photo, 0, 0, overlay.width, overlay.height);
+ }
+ if (state.photoDetections.length) drawPhotoDetections();
}
for (const [trackId, track] of state.tracks) {
@@ -186,6 +209,24 @@ function render() {
ctx.globalAlpha = 1;
}
+ if (state.mode === 'live' || state.mode === 'photo') {
+ const tagSource = state.mode === 'live'
+ ? [...state.tracks.values()].map((track) => track.identity && track.priceTag).filter(Boolean)
+ : state.photoDetections.map((d) => d.price_tag).filter(Boolean);
+ for (const tag of tagSource) {
+ if (!tag.box) continue;
+ const readable = tag.state === 'readable';
+ drawBox(tag.box, {
+ color: readable ? '#0ea5e9' : 'rgba(14,165,233,0.5)',
+ width: readable ? 2 : 1,
+ dashed: !readable,
+ });
+ if (readable && tag.price !== null && tag.price !== undefined) {
+ chip(`¥${tag.price}`, tag.box[0], tag.box[1], '#0ea5e9', '#ffffff');
+ }
+ }
+ }
+
if (state.showEmpty && state.mode === 'live') {
for (const slot of state.slots) {
drawBox(slot.box, { color: '#ef4444', width: 3, fill: 'rgba(239,68,68,0.20)' });
@@ -421,6 +462,29 @@ function sendFrame() {
}, 'image/jpeg', state.quality);
}
+function renderCompliance(compliance) {
+ const strip = el('complianceStrip');
+ if (!compliance) { strip.hidden = true; return; }
+ const rows = compliance.rows || [];
+ const items = rows.slice(0, 3).map(
+ (row) => `⚠ ${t('rule_' + row.rule)} · ${t('row')} ${row.row_index} (${row.missing})`
+ );
+ if (rows.length > 3) {
+ items.push(`+${rows.length - 3} ${t('more')}`);
+ }
+ const hints = compliance.hints || [];
+ if (!items.length && hints.length) {
+ strip.hidden = false;
+ strip.dataset.kind = 'hint';
+ strip.innerHTML = hints.map((hint) => `${hint.detail}`).join('');
+ return;
+ }
+ if (!items.length) { strip.hidden = true; return; }
+ strip.hidden = false;
+ strip.dataset.kind = 'violation';
+ strip.innerHTML = items.join('');
+}
+
function handleResult(message) {
state.results += 1;
state.inFlight = false;
@@ -437,6 +501,8 @@ function handleResult(message) {
box: existing ? existing.box : [...track.box],
target: track.box,
identity: track.identity,
+ priceTag: track.price_tag || null,
+ reconciliation: track.reconciliation || null,
product: track.product || productFor(track.identity.sku_id),
lastSeen: performance.now(),
});
@@ -455,6 +521,7 @@ function handleResult(message) {
else if (hint === 'too_far') showHint('warning', t('hintTooFar'));
else showHint(null, '');
}
+ renderCompliance(message.summary && message.summary.compliance);
renderInView();
if (state.selectedTrack !== null && !state.cardPinned) {
const track = state.tracks.get(state.selectedTrack);
@@ -494,12 +561,12 @@ async function recognizeBlob(blob, previewUrl) {
setStatus('neutral', t('identifying'));
const photo = el('photo');
photo.src = previewUrl || URL.createObjectURL(blob);
- photo.hidden = false;
+ photo.hidden = true; // the canvas draws it
el('video').hidden = true;
el('stageEmpty').hidden = true;
await photo.decode().catch(() => {});
- const response = await fetch(api('/v1/recognize?classify=false&empty_slots=true'), {
+ const response = await fetch(api('/v1/recognize?classify=false&empty_slots=true&tags=true'), {
method: 'POST',
headers: { 'Content-Type': 'image/jpeg', ...headers() },
body: blob,
@@ -514,8 +581,13 @@ async function recognizeBlob(blob, previewUrl) {
state.photoWidth = data.width;
state.photoHeight = data.height;
state.slots = data.empty_slots || [];
- setCanvasSize(photo.naturalWidth || data.width, photo.naturalHeight || data.height);
+ const cap = 2048;
+ const naturalWidth = photo.naturalWidth || data.width;
+ const naturalHeight = photo.naturalHeight || data.height;
+ const scale = Math.min(1, cap / Math.max(1, naturalWidth));
+ setCanvasSize(Math.round(naturalWidth * scale), Math.round(naturalHeight * scale));
el('inViewCount').textContent = String(data.detections.length);
+ renderCompliance(data.compliance);
setStatus('neutral', t('detected'));
classifyPhoto();
}
@@ -530,11 +602,16 @@ async function classifyPhoto(detectionIds = null) {
});
if (!response.ok) { setStatus('danger', t('noServer')); return; }
const data = await response.json();
- const byId = new Map(data.identities.map((row) => [row.detection_id, row.identity]));
+ const byId = new Map(data.identities.map((row) => [row.detection_id, row]));
for (const detection of state.photoDetections) {
- if (byId.has(detection.id)) detection.identity = byId.get(detection.id);
+ const row = byId.get(detection.id);
+ if (!row) continue;
+ detection.identity = row.identity;
+ if (row.reconciliation) detection.reconciliation = row.reconciliation;
+ if (row.price_tag) detection.price_tag = row.price_tag;
}
const summary = data.summary || {};
+ renderCompliance(data.compliance);
setStatus('recognized', `${summary.recognized || 0} / ${data.identities.length}`);
renderInView();
}
@@ -601,6 +678,8 @@ function selectBySku(skuId) {
function showTrackCard(track) {
const identity = track.identity || {};
+ state.currentTag = track.priceTag || null;
+ state.currentReconciliation = track.reconciliation || null;
if (identity.state === 'recognized') {
const product = track.product || productFor(identity.sku_id);
showProductCard(product, identity);
@@ -613,6 +692,34 @@ function showTrackCard(track) {
}
}
+function renderOcrSection(tag, reconciliation) {
+ const section = el('ocrSection');
+ const chipEl = el('reconChip');
+ const rows = el('ocrRows');
+ if (!tag && !reconciliation) { section.hidden = true; return; }
+ section.hidden = false;
+ const state = (reconciliation || {}).state;
+ chipEl.dataset.state = state || 'unverified';
+ chipEl.textContent = t(`recon_${state || 'unverified'}`);
+ const lines = [];
+ if (tag) {
+ if (tag.price !== null && tag.price !== undefined) {
+ const bad = state === 'mismatch';
+ lines.push([t('ocrPrice'), `¥${tag.price}`, bad]);
+ }
+ if (tag.regular_price) lines.push([t('ocrRegular'), `¥${tag.regular_price}`, false]);
+ if (tag.discount) lines.push([t('ocrDiscount'), `${Math.round(tag.discount * 100)}%`, false]);
+ if (tag.unit_price) lines.push([t('ocrUnit'), `${tag.unit_price}`, false]);
+ if (tag.name) lines.push([t('ocrName'), tag.name, false]);
+ if (tag.state === 'unreadable') lines.push([t('ocrState'), t('ocrUnreadable'), true]);
+ if (tag.state === 'absent') lines.push([t('ocrState'), t('ocrAbsent'), false]);
+ }
+ if (reconciliation && reconciliation.message) lines.push([t('ocrNote'), reconciliation.message, state === 'mismatch']);
+ rows.innerHTML = lines
+ .map(([key, value, bad]) => `${key}${value}
`)
+ .join('');
+}
+
function showProductCard(product, identity) {
if (!product) { showPlainCard(t('detected'), ''); return; }
el('cardPlaceholder').hidden = true;
@@ -631,6 +738,7 @@ function showProductCard(product, identity) {
el('showMatchBtn').disabled = false;
const confidence = identity.confidence ? ` · ${(identity.confidence * 100).toFixed(0)}%` : '';
el('scoreLabel').textContent = identity.score ? `score ${identity.score}${confidence}` : '';
+ renderOcrSection(state.currentTag, state.currentReconciliation);
setStatus('recognized', t('recognized'));
}
@@ -655,6 +763,7 @@ function showCandidatesCard(identity) {
candidates.innerHTML = `${t('candidatesTitle')}
${rows}`;
candidates.hidden = false;
el('showMatchBtn').disabled = false;
+ renderOcrSection(state.currentTag, state.currentReconciliation);
setStatus('needs', t('needsConfirm'));
}
@@ -672,6 +781,7 @@ function showPlainCard(title, note) {
el('metricExpiry').textContent = '—';
el('candidates').hidden = true;
el('showMatchBtn').disabled = true;
+ renderOcrSection(state.currentTag, state.currentReconciliation);
setStatus('danger', title);
if (note) el('productPrice').textContent = note;
}
@@ -705,7 +815,10 @@ overlay.addEventListener('click', (event) => {
return;
}
if (state.mode === 'photo') {
- const [px, py] = photoPointToDetectionSpace(x, y);
+ // The canvas backing store matches the server's pixel space (both scaled
+ // from the same image), so no extra transform is needed.
+ const px = x * (state.photoWidth / Math.max(1, overlay.width));
+ const py = y * (state.photoHeight / Math.max(1, overlay.height));
let best = null;
for (const detection of state.photoDetections) {
const [x1, y1, x2, y2] = detection.box;
@@ -735,6 +848,8 @@ function photoPointToDetectionSpace(x, y) {
function selectPhotoDetection(detection) {
state.selectedDetection = detection.id;
state.selectedTrack = null;
+ state.currentTag = detection.price_tag || null;
+ state.currentReconciliation = detection.reconciliation || null;
const identity = detection.identity || {};
if (identity.state === 'recognized') showProductCard(productFor(identity.sku_id), identity);
else if (identity.state === 'needs_confirmation') showCandidatesCard(identity);
@@ -813,7 +928,7 @@ function setMode(mode) {
el('liveControls').hidden = mode !== 'live';
el('photoControls').hidden = mode === 'live';
el('video').hidden = mode !== 'live';
- el('photo').hidden = mode !== 'photo';
+ el('photo').hidden = true;
state.tracks.clear();
state.slots = [];
state.photoDetections = [];
diff --git a/recognization/code/web/index.html b/recognization/code/web/index.html
index c827f80..8e96ea5 100644
--- a/recognization/code/web/index.html
+++ b/recognization/code/web/index.html
@@ -87,6 +87,7 @@
Start the live scan, or open Photo mode to inspect a shelf image.
+
@@ -128,6 +129,13 @@
+
diff --git a/recognization/code/web/styles.css b/recognization/code/web/styles.css
index 2418236..acbcabc 100644
--- a/recognization/code/web/styles.css
+++ b/recognization/code/web/styles.css
@@ -130,13 +130,20 @@ main {
background: var(--night);
border-radius: 14px;
overflow: hidden;
- min-height: 540px;
+ height: 70vh;
+ min-height: 420px;
display: flex;
align-items: center;
justify-content: center;
}
-.stage video, .stage img { max-width: 100%; max-height: 74vh; display: block; }
-.stage canvas { position: absolute; inset: 0; width: 100%; height: 100%; }
+.stage video, .stage img { max-width: 100%; max-height: 100%; display: block; }
+/* The canvas is the same contained box as the media it annotates: its backing
+ store is the media's own pixel size, so server coordinates map 1:1 and the
+ element rect (used for click mapping) is exactly the visible picture. */
+.stage canvas {
+ position: absolute; left: 50%; top: 50%; transform: translate(-50%, -50%);
+ max-width: 100%; max-height: 100%;
+}
.stage-empty { position: absolute; inset: 0; display: flex; flex-direction: column; align-items: center; justify-content: center; gap: 10px; color: #cbd5e1; font-size: 13px; text-align: center; padding: 24px; }
.stage-empty-icon { font-size: 34px; }
@@ -220,5 +227,27 @@ main {
@media (max-width: 1024px) {
main { grid-template-columns: 1fr; padding: 16px; }
- .stage { min-height: 380px; }
+ .stage { height: 56vh; min-height: 320px; }
}
+
+.ocr-section { border-top: 1px solid var(--border-color); padding-top: 12px; display: flex; flex-direction: column; gap: 8px; }
+.ocr-header { display: flex; justify-content: space-between; align-items: center; font-size: 11px; font-weight: 800; color: #0369a1; text-transform: uppercase; letter-spacing: 0.5px; }
+.recon-chip { font-size: 10px; font-weight: 800; padding: 2px 8px; border-radius: 999px; text-transform: none; letter-spacing: 0; }
+.recon-chip[data-state="confirmed"] { background: var(--primary-light); color: var(--primary-dark); }
+.recon-chip[data-state="mismatch"] { background: var(--danger-soft); color: #b91c1c; }
+.recon-chip[data-state="unverified"], .recon-chip[data-state="no_tag"] { background: var(--slate-soft); color: var(--text-muted); }
+.recon-chip[data-state="tag_only"] { background: #e0f2fe; color: #0369a1; }
+.ocr-rows { display: flex; flex-direction: column; gap: 4px; }
+.ocr-row { display: flex; justify-content: space-between; gap: 10px; font-size: 12px; }
+.ocr-row .ocr-key { color: var(--text-muted); font-weight: 600; }
+.ocr-row .ocr-val { color: var(--text-main); font-weight: 700; text-align: right; }
+.ocr-row .ocr-val.bad { color: #b91c1c; }
+.compliance-strip {
+ position: absolute; top: 14px; left: 14px; right: 14px;
+ background: rgba(255, 255, 255, 0.94); backdrop-filter: blur(8px);
+ border: 1px solid #fca5a5; border-radius: 12px; padding: 8px 14px;
+ font-size: 12px; font-weight: 700; color: #b91c1c; display: flex; gap: 12px; flex-wrap: wrap;
+ box-shadow: 0 4px 16px rgba(0, 0, 0, 0.08); pointer-events: none;
+}
+.compliance-strip[data-kind="hint"] { border-color: var(--amber); color: #92400e; }
+.compliance-strip .compliance-item { display: inline-flex; align-items: center; gap: 6px; }