Spaces:
Runtime error
Runtime error
Rajeev Pandey commited on
Commit ·
e1242e7
1
Parent(s): 12a86e2
Improve document scrolling and interactive insights
Browse files- TASK_SHEET.md +7 -1
- backend/document_service.py +66 -28
- backend/glossary_service.py +5 -1
- data/project_progress.json +1 -1
- frontend/camera.js +3 -3
- frontend/overlay.js +44 -0
- frontend/styles.css +4 -2
- tests/test_services.py +38 -0
TASK_SHEET.md
CHANGED
|
@@ -93,6 +93,7 @@ Every item below must pass before FalconScan is marked ready for GitHub or Huggi
|
|
| 93 |
- [x] No more than three highest-confidence dots are shown on the page at once.
|
| 94 |
- [x] Dots use the reduced 18px control size and can be dragged to uncover document text.
|
| 95 |
- [x] A readable single-word OCR region can receive a contextual dot when it is not already represented by a verified term.
|
|
|
|
| 96 |
|
| 97 |
### Selection, highlight, and business insight
|
| 98 |
|
|
@@ -118,6 +119,8 @@ Every item below must pass before FalconScan is marked ready for GitHub or Huggi
|
|
| 118 |
- [x] Files over 15 MB are rejected before analysis.
|
| 119 |
- [x] Uploaded document images and files are not persisted by default.
|
| 120 |
- [x] Uploaded document previews use a dedicated scroll container with visible native scrollbars.
|
|
|
|
|
|
|
| 121 |
|
| 122 |
### Camera and performance
|
| 123 |
|
|
@@ -129,6 +132,8 @@ Every item below must pass before FalconScan is marked ready for GitHub or Huggi
|
|
| 129 |
- [x] Large uploaded images are resized to a maximum 1600px edge in the browser before transfer.
|
| 130 |
- [x] Small images avoid unnecessary recompression.
|
| 131 |
- [x] Portable CPU OCR is warmed in the background to reduce first-scan latency.
|
|
|
|
|
|
|
| 132 |
|
| 133 |
### Responsive UI and accessibility
|
| 134 |
|
|
@@ -151,10 +156,11 @@ Every item below must pass before FalconScan is marked ready for GitHub or Huggi
|
|
| 151 |
- [x] SME-approved definitions override pending user corrections.
|
| 152 |
- [x] Feedback and review actions preserve an audit history.
|
| 153 |
- [x] AI-generated or unmatched explanations are marked unverified.
|
|
|
|
| 154 |
|
| 155 |
### Automated verification recorded
|
| 156 |
|
| 157 |
-
- [x] Python/API suite:
|
| 158 |
- [x] JavaScript syntax checks: camera, overlay, feedback, and insights scripts passing.
|
| 159 |
- [x] DOM integration: two terms produced two dots.
|
| 160 |
- [x] DOM integration: dot click expanded the inline definition.
|
|
|
|
| 93 |
- [x] No more than three highest-confidence dots are shown on the page at once.
|
| 94 |
- [x] Dots use the reduced 18px control size and can be dragged to uncover document text.
|
| 95 |
- [x] A readable single-word OCR region can receive a contextual dot when it is not already represented by a verified term.
|
| 96 |
+
- [x] Clicking selectable document text creates a dot at the clicked location without exceeding the three-dot limit.
|
| 97 |
|
| 98 |
### Selection, highlight, and business insight
|
| 99 |
|
|
|
|
| 119 |
- [x] Files over 15 MB are rejected before analysis.
|
| 120 |
- [x] Uploaded document images and files are not persisted by default.
|
| 121 |
- [x] Uploaded document previews use a dedicated scroll container with visible native scrollbars.
|
| 122 |
+
- [x] Multi-page PDFs render into one continuous scroll surface with positioned text and dots across pages.
|
| 123 |
+
- [x] DOCX rendering grows with document content rather than stopping at the earlier fixed preview height.
|
| 124 |
|
| 125 |
### Camera and performance
|
| 126 |
|
|
|
|
| 132 |
- [x] Large uploaded images are resized to a maximum 1600px edge in the browser before transfer.
|
| 133 |
- [x] Small images avoid unnecessary recompression.
|
| 134 |
- [x] Portable CPU OCR is warmed in the background to reduce first-scan latency.
|
| 135 |
+
- [x] Uploaded images are preprocessed for contrast/sharpness at a bounded 1400px edge.
|
| 136 |
+
- [x] The last four document analyses are cached in memory for immediate repeat uploads.
|
| 137 |
|
| 138 |
### Responsive UI and accessibility
|
| 139 |
|
|
|
|
| 156 |
- [x] SME-approved definitions override pending user corrections.
|
| 157 |
- [x] Feedback and review actions preserve an audit history.
|
| 158 |
- [x] AI-generated or unmatched explanations are marked unverified.
|
| 159 |
+
- [x] Confidence combines recognition and knowledge signals using calibrated weighting rather than artificially multiplying scores downward.
|
| 160 |
|
| 161 |
### Automated verification recorded
|
| 162 |
|
| 163 |
+
- [x] Python/API suite: 15 tests passing.
|
| 164 |
- [x] JavaScript syntax checks: camera, overlay, feedback, and insights scripts passing.
|
| 165 |
- [x] DOM integration: two terms produced two dots.
|
| 166 |
- [x] DOM integration: dot click expanded the inline definition.
|
backend/document_service.py
CHANGED
|
@@ -1,11 +1,13 @@
|
|
| 1 |
import base64
|
|
|
|
| 2 |
import io
|
| 3 |
import textwrap
|
| 4 |
import zipfile
|
|
|
|
| 5 |
from pathlib import Path
|
| 6 |
from xml.etree import ElementTree
|
| 7 |
|
| 8 |
-
from PIL import Image, ImageDraw, ImageFont
|
| 9 |
|
| 10 |
from backend.glossary_service import GlossaryService
|
| 11 |
from backend.ocr_service import OCRService
|
|
@@ -13,10 +15,12 @@ from backend.ocr_service import OCRService
|
|
| 13 |
|
| 14 |
class DocumentService:
|
| 15 |
MAX_BYTES = 15 * 1024 * 1024
|
|
|
|
| 16 |
|
| 17 |
def __init__(self, ocr: OCRService, glossary: GlossaryService):
|
| 18 |
self.ocr = ocr
|
| 19 |
self.glossary = glossary
|
|
|
|
| 20 |
|
| 21 |
def analyze(self, encoded: str, filename: str, language: str) -> dict:
|
| 22 |
try:
|
|
@@ -25,14 +29,23 @@ class DocumentService:
|
|
| 25 |
raise ValueError("Invalid uploaded document") from exc
|
| 26 |
if not raw or len(raw) > self.MAX_BYTES:
|
| 27 |
raise ValueError("Document must be between 1 byte and 15 MB")
|
|
|
|
|
|
|
|
|
|
|
|
|
| 28 |
extension = Path(filename).suffix.lower()
|
| 29 |
if extension in {".jpg", ".jpeg", ".png", ".webp"}:
|
| 30 |
-
|
| 31 |
-
|
| 32 |
-
|
| 33 |
-
|
| 34 |
-
|
| 35 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 36 |
|
| 37 |
def _image(self, raw: bytes, language: str) -> dict:
|
| 38 |
try:
|
|
@@ -40,7 +53,9 @@ class DocumentService:
|
|
| 40 |
image.load()
|
| 41 |
except Exception as exc:
|
| 42 |
raise ValueError("The uploaded image could not be opened") from exc
|
| 43 |
-
|
|
|
|
|
|
|
| 44 |
terms = self.glossary.match_regions(detections)
|
| 45 |
if len(terms) < 3:
|
| 46 |
terms.extend(self.glossary.contextual_fallbacks(detections, 3 - len(terms), terms))
|
|
@@ -56,28 +71,51 @@ class DocumentService:
|
|
| 56 |
document = fitz.open(stream=raw, filetype="pdf")
|
| 57 |
if document.page_count < 1:
|
| 58 |
raise ValueError("The PDF has no pages")
|
| 59 |
-
|
| 60 |
-
|
| 61 |
-
|
| 62 |
-
|
| 63 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 64 |
regions = []
|
| 65 |
-
|
| 66 |
-
|
| 67 |
-
|
| 68 |
-
|
| 69 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 70 |
terms = self.glossary.match_regions(regions)
|
| 71 |
if len(terms) < 3:
|
| 72 |
terms.extend(self.glossary.contextual_fallbacks(regions, 3 - len(terms), terms))
|
| 73 |
-
|
| 74 |
-
|
| 75 |
-
|
| 76 |
-
|
| 77 |
-
terms.extend(self.glossary.contextual_fallbacks(detections, 3 - len(terms), terms))
|
| 78 |
-
_, unknown = self.glossary.match_ocr(detections)
|
| 79 |
-
return self._result(image, terms, detections, unknown, "scanned_pdf_ocr")
|
| 80 |
-
return self._result(image, terms, regions, [], "pdf_text")
|
| 81 |
except (ValueError, RuntimeError):
|
| 82 |
raise
|
| 83 |
except Exception as exc:
|
|
@@ -105,7 +143,7 @@ class DocumentService:
|
|
| 105 |
for paragraph in paragraphs:
|
| 106 |
wrapped.extend(textwrap.wrap(paragraph, width=85, break_long_words=False) or [""])
|
| 107 |
wrapped.append("")
|
| 108 |
-
height =
|
| 109 |
image = Image.new("RGB", (width, height), "#ffffff")
|
| 110 |
draw = ImageDraw.Draw(image)
|
| 111 |
try:
|
|
@@ -134,7 +172,7 @@ class DocumentService:
|
|
| 134 |
@staticmethod
|
| 135 |
def _result(image: Image.Image, terms: list, detections: list, unknown: list, method: str) -> dict:
|
| 136 |
output = io.BytesIO()
|
| 137 |
-
image.save(output, format="JPEG", quality=
|
| 138 |
return {
|
| 139 |
"detected_terms": terms,
|
| 140 |
"ocr_items": detections,
|
|
|
|
| 1 |
import base64
|
| 2 |
+
import hashlib
|
| 3 |
import io
|
| 4 |
import textwrap
|
| 5 |
import zipfile
|
| 6 |
+
from collections import OrderedDict
|
| 7 |
from pathlib import Path
|
| 8 |
from xml.etree import ElementTree
|
| 9 |
|
| 10 |
+
from PIL import Image, ImageDraw, ImageFilter, ImageFont, ImageOps
|
| 11 |
|
| 12 |
from backend.glossary_service import GlossaryService
|
| 13 |
from backend.ocr_service import OCRService
|
|
|
|
| 15 |
|
| 16 |
class DocumentService:
|
| 17 |
MAX_BYTES = 15 * 1024 * 1024
|
| 18 |
+
MAX_PDF_PAGES = 12
|
| 19 |
|
| 20 |
def __init__(self, ocr: OCRService, glossary: GlossaryService):
|
| 21 |
self.ocr = ocr
|
| 22 |
self.glossary = glossary
|
| 23 |
+
self._cache: OrderedDict[str, dict] = OrderedDict()
|
| 24 |
|
| 25 |
def analyze(self, encoded: str, filename: str, language: str) -> dict:
|
| 26 |
try:
|
|
|
|
| 29 |
raise ValueError("Invalid uploaded document") from exc
|
| 30 |
if not raw or len(raw) > self.MAX_BYTES:
|
| 31 |
raise ValueError("Document must be between 1 byte and 15 MB")
|
| 32 |
+
cache_key = hashlib.sha256(raw + filename.lower().encode() + language.encode()).hexdigest()
|
| 33 |
+
if cache_key in self._cache:
|
| 34 |
+
self._cache.move_to_end(cache_key)
|
| 35 |
+
return {**self._cache[cache_key], "cached": True}
|
| 36 |
extension = Path(filename).suffix.lower()
|
| 37 |
if extension in {".jpg", ".jpeg", ".png", ".webp"}:
|
| 38 |
+
result = self._image(raw, language)
|
| 39 |
+
elif extension == ".pdf":
|
| 40 |
+
result = self._pdf(raw)
|
| 41 |
+
elif extension == ".docx":
|
| 42 |
+
result = self._docx(raw)
|
| 43 |
+
else:
|
| 44 |
+
raise ValueError("Supported formats are JPG, PNG, WebP, PDF, and DOCX")
|
| 45 |
+
self._cache[cache_key] = result
|
| 46 |
+
while len(self._cache) > 4:
|
| 47 |
+
self._cache.popitem(last=False)
|
| 48 |
+
return {**result, "cached": False}
|
| 49 |
|
| 50 |
def _image(self, raw: bytes, language: str) -> dict:
|
| 51 |
try:
|
|
|
|
| 53 |
image.load()
|
| 54 |
except Exception as exc:
|
| 55 |
raise ValueError("The uploaded image could not be opened") from exc
|
| 56 |
+
image.thumbnail((1400, 1400), Image.Resampling.LANCZOS)
|
| 57 |
+
prepared = ImageOps.autocontrast(image.convert("L")).filter(ImageFilter.SHARPEN).convert("RGB")
|
| 58 |
+
detections = self.ocr.extract(prepared, language)
|
| 59 |
terms = self.glossary.match_regions(detections)
|
| 60 |
if len(terms) < 3:
|
| 61 |
terms.extend(self.glossary.contextual_fallbacks(detections, 3 - len(terms), terms))
|
|
|
|
| 71 |
document = fitz.open(stream=raw, filetype="pdf")
|
| 72 |
if document.page_count < 1:
|
| 73 |
raise ValueError("The PDF has no pages")
|
| 74 |
+
page_count = min(document.page_count, self.MAX_PDF_PAGES)
|
| 75 |
+
rendered = []
|
| 76 |
+
canvas_width = 0
|
| 77 |
+
for index in range(page_count):
|
| 78 |
+
page = document[index]
|
| 79 |
+
zoom = min(1.5, 1100 / max(1, page.rect.width))
|
| 80 |
+
pixmap = page.get_pixmap(matrix=fitz.Matrix(zoom, zoom), alpha=False)
|
| 81 |
+
page_image = Image.frombytes("RGB", [pixmap.width, pixmap.height], pixmap.samples)
|
| 82 |
+
rendered.append((page, page_image, zoom))
|
| 83 |
+
canvas_width = max(canvas_width, page_image.width)
|
| 84 |
+
gap = 18
|
| 85 |
+
total_height = sum(item[1].height for item in rendered) + gap * max(0, page_count - 1)
|
| 86 |
+
image = Image.new("RGB", (canvas_width, total_height), "#dfe3df")
|
| 87 |
regions = []
|
| 88 |
+
y_offset = 0
|
| 89 |
+
for page, page_image, zoom in rendered:
|
| 90 |
+
x_offset = (canvas_width - page_image.width) // 2
|
| 91 |
+
image.paste(page_image, (x_offset, y_offset))
|
| 92 |
+
page_regions = []
|
| 93 |
+
for block in page.get_text("blocks"):
|
| 94 |
+
x1, y1, x2, y2, text = block[:5]
|
| 95 |
+
if text.strip():
|
| 96 |
+
page_regions.append({
|
| 97 |
+
"text": text,
|
| 98 |
+
"bbox": [x_offset + x1 * zoom, y_offset + y1 * zoom,
|
| 99 |
+
x_offset + x2 * zoom, y_offset + y2 * zoom],
|
| 100 |
+
"confidence": 1.0,
|
| 101 |
+
"language": self._language(text),
|
| 102 |
+
})
|
| 103 |
+
if page_regions:
|
| 104 |
+
regions.extend(page_regions)
|
| 105 |
+
else:
|
| 106 |
+
page_detections = self.ocr.extract(page_image, "en")
|
| 107 |
+
for item in page_detections:
|
| 108 |
+
item["bbox"] = [item["bbox"][0] + x_offset, item["bbox"][1] + y_offset,
|
| 109 |
+
item["bbox"][2] + x_offset, item["bbox"][3] + y_offset]
|
| 110 |
+
regions.extend(page_detections)
|
| 111 |
+
y_offset += page_image.height + gap
|
| 112 |
terms = self.glossary.match_regions(regions)
|
| 113 |
if len(terms) < 3:
|
| 114 |
terms.extend(self.glossary.contextual_fallbacks(regions, 3 - len(terms), terms))
|
| 115 |
+
result = self._result(image, terms, regions, [], "pdf_multpage")
|
| 116 |
+
result.update({"page_count": document.page_count, "pages_analyzed": page_count,
|
| 117 |
+
"truncated": document.page_count > page_count})
|
| 118 |
+
return result
|
|
|
|
|
|
|
|
|
|
|
|
|
| 119 |
except (ValueError, RuntimeError):
|
| 120 |
raise
|
| 121 |
except Exception as exc:
|
|
|
|
| 143 |
for paragraph in paragraphs:
|
| 144 |
wrapped.extend(textwrap.wrap(paragraph, width=85, break_long_words=False) or [""])
|
| 145 |
wrapped.append("")
|
| 146 |
+
height = max(900, margin * 2 + len(wrapped) * line_height)
|
| 147 |
image = Image.new("RGB", (width, height), "#ffffff")
|
| 148 |
draw = ImageDraw.Draw(image)
|
| 149 |
try:
|
|
|
|
| 172 |
@staticmethod
|
| 173 |
def _result(image: Image.Image, terms: list, detections: list, unknown: list, method: str) -> dict:
|
| 174 |
output = io.BytesIO()
|
| 175 |
+
image.save(output, format="JPEG", quality=80, optimize=True)
|
| 176 |
return {
|
| 177 |
"detected_terms": terms,
|
| 178 |
"ocr_items": detections,
|
backend/glossary_service.py
CHANGED
|
@@ -144,8 +144,12 @@ class GlossaryService:
|
|
| 144 |
"ocr_confidence": region.get("confidence", 1.0),
|
| 145 |
"language": region.get("language", "en"),
|
| 146 |
})
|
|
|
|
|
|
|
|
|
|
|
|
|
| 147 |
definition["confidence"] = round(
|
| 148 |
-
|
| 149 |
)
|
| 150 |
found.append(definition)
|
| 151 |
return found
|
|
|
|
| 144 |
"ocr_confidence": region.get("confidence", 1.0),
|
| 145 |
"language": region.get("language", "en"),
|
| 146 |
})
|
| 147 |
+
knowledge_confidence = float(definition["confidence"])
|
| 148 |
+
recognition_confidence = float(region.get("confidence", 1.0))
|
| 149 |
+
# Calibrate two independent signals instead of multiplying them into an
|
| 150 |
+
# artificially low score. OCR remains the dominant factor.
|
| 151 |
definition["confidence"] = round(
|
| 152 |
+
0.72 * recognition_confidence + 0.28 * knowledge_confidence, 3
|
| 153 |
)
|
| 154 |
found.append(definition)
|
| 155 |
return found
|
data/project_progress.json
CHANGED
|
@@ -5,7 +5,7 @@
|
|
| 5 |
"current_module": "Representative device validation",
|
| 6 |
"pending_modules": ["Install production OCR models", "Configure optional hosted AI endpoint", "Configure optional VLM endpoint", "Persistent production database"],
|
| 7 |
"open_issues": ["Local JSON storage is single-instance and ephemeral on default free Spaces", "PaddleOCR model download adds cold-start time", "Remote AI and VLM adapters require provider-specific endpoint contracts"],
|
| 8 |
-
"last_completed_step": "Added
|
| 9 |
"next_recommended_step": "Run the remaining physical iPhone/Android acceptance checks, then execute GitHub CI and Hugging Face smoke tests after remotes are configured",
|
| 10 |
"last_updated": "2026-06-20T12:00:00Z"
|
| 11 |
}
|
|
|
|
| 5 |
"current_module": "Representative device validation",
|
| 6 |
"pending_modules": ["Install production OCR models", "Configure optional hosted AI endpoint", "Configure optional VLM endpoint", "Persistent production database"],
|
| 7 |
"open_issues": ["Local JSON storage is single-instance and ephemeral on default free Spaces", "PaddleOCR model download adds cold-start time", "Remote AI and VLM adapters require provider-specific endpoint contracts"],
|
| 8 |
+
"last_completed_step": "Added continuous multi-page scrolling, click-text-to-dot interaction, calibrated confidence, OCR preprocessing, and repeat-upload caching",
|
| 9 |
"next_recommended_step": "Run the remaining physical iPhone/Android acceptance checks, then execute GitHub CI and Hugging Face smoke tests after remotes are configured",
|
| 10 |
"last_updated": "2026-06-20T12:00:00Z"
|
| 11 |
}
|
frontend/camera.js
CHANGED
|
@@ -218,16 +218,16 @@
|
|
| 218 |
const objectUrl = URL.createObjectURL(file);
|
| 219 |
try {
|
| 220 |
const image = await loadImage(objectUrl);
|
| 221 |
-
const maxEdge =
|
| 222 |
const scale = Math.min(1, maxEdge / Math.max(image.naturalWidth, image.naturalHeight));
|
| 223 |
-
if (scale === 1 && file.size < 2
|
| 224 |
return { base64: await readBase64(file), filename: file.name };
|
| 225 |
}
|
| 226 |
const canvas = document.createElement("canvas");
|
| 227 |
canvas.width = Math.max(1, Math.round(image.naturalWidth * scale));
|
| 228 |
canvas.height = Math.max(1, Math.round(image.naturalHeight * scale));
|
| 229 |
canvas.getContext("2d").drawImage(image, 0, 0, canvas.width, canvas.height);
|
| 230 |
-
const blob = await new Promise((resolve) => canvas.toBlob(resolve, "image/jpeg", 0.
|
| 231 |
return { base64: await readBase64(blob), filename: "optimized-upload.jpg" };
|
| 232 |
} finally {
|
| 233 |
URL.revokeObjectURL(objectUrl);
|
|
|
|
| 218 |
const objectUrl = URL.createObjectURL(file);
|
| 219 |
try {
|
| 220 |
const image = await loadImage(objectUrl);
|
| 221 |
+
const maxEdge = 1400;
|
| 222 |
const scale = Math.min(1, maxEdge / Math.max(image.naturalWidth, image.naturalHeight));
|
| 223 |
+
if (scale === 1 && file.size < 2 * 1024 * 1024) {
|
| 224 |
return { base64: await readBase64(file), filename: file.name };
|
| 225 |
}
|
| 226 |
const canvas = document.createElement("canvas");
|
| 227 |
canvas.width = Math.max(1, Math.round(image.naturalWidth * scale));
|
| 228 |
canvas.height = Math.max(1, Math.round(image.naturalHeight * scale));
|
| 229 |
canvas.getContext("2d").drawImage(image, 0, 0, canvas.width, canvas.height);
|
| 230 |
+
const blob = await new Promise((resolve) => canvas.toBlob(resolve, "image/jpeg", 0.80));
|
| 231 |
return { base64: await readBase64(blob), filename: "optimized-upload.jpg" };
|
| 232 |
} finally {
|
| 233 |
URL.revokeObjectURL(objectUrl);
|
frontend/overlay.js
CHANGED
|
@@ -114,6 +114,49 @@ window.FalconOverlay = (() => {
|
|
| 114 |
marker.addEventListener("pointercancel", finish);
|
| 115 |
}
|
| 116 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 117 |
function openPopover(item, marker, layout) {
|
| 118 |
closePopover();
|
| 119 |
marker.classList.add("active");
|
|
@@ -160,6 +203,7 @@ window.FalconOverlay = (() => {
|
|
| 160 |
span.style.width = `${Math.max(8, (x2 - x1) * layout.scale)}px`;
|
| 161 |
span.style.height = `${Math.max(12, (y2 - y1) * layout.scale)}px`;
|
| 162 |
span.style.fontSize = `${clamp((y2 - y1) * layout.scale * 0.72, 8, 24)}px`;
|
|
|
|
| 163 |
layer.appendChild(span);
|
| 164 |
});
|
| 165 |
const explainSelection = () => {
|
|
|
|
| 114 |
marker.addEventListener("pointercancel", finish);
|
| 115 |
}
|
| 116 |
|
| 117 |
+
function triggerDotAtText(event, region, layout) {
|
| 118 |
+
const selection = window.getSelection();
|
| 119 |
+
if (selection && !selection.isCollapsed && selection.toString().trim()) return;
|
| 120 |
+
const existingMarkers = [...overlay.querySelectorAll(".term-marker")];
|
| 121 |
+
if (existingMarkers.length >= 3) {
|
| 122 |
+
(overlay.querySelector(".term-marker.user-triggered") || existingMarkers[existingMarkers.length - 1]).remove();
|
| 123 |
+
}
|
| 124 |
+
const match = terms.find((item) =>
|
| 125 |
+
item.bbox?.every((value, index) => Math.abs(value - region.bbox[index]) < 2)
|
| 126 |
+
|| String(region.text).toLowerCase().includes(String(item.term).toLowerCase())
|
| 127 |
+
);
|
| 128 |
+
const item = match || {
|
| 129 |
+
term: String(region.text).trim().slice(0, 58),
|
| 130 |
+
definition: "Select this phrase for a contextual summary and business meaning.",
|
| 131 |
+
confidence: Math.min(.72, Number(region.confidence || .75) * .72),
|
| 132 |
+
source: "ai_generated_unverified",
|
| 133 |
+
source_label: "Document context · needs verification",
|
| 134 |
+
category: "Document Context",
|
| 135 |
+
related_terms: [],
|
| 136 |
+
document_types: [],
|
| 137 |
+
bbox: region.bbox,
|
| 138 |
+
};
|
| 139 |
+
const overlayRect = overlay.getBoundingClientRect();
|
| 140 |
+
const marker = document.createElement("button");
|
| 141 |
+
marker.className = "term-marker user-triggered";
|
| 142 |
+
marker.type = "button";
|
| 143 |
+
marker.style.left = `${clamp(event.clientX - overlayRect.left - 9, 5, layout.viewRect.width - 23)}px`;
|
| 144 |
+
marker.style.top = `${clamp(event.clientY - overlayRect.top - 9, 5, layout.viewRect.height - 23)}px`;
|
| 145 |
+
marker.setAttribute("aria-label", `Explain ${item.term}`);
|
| 146 |
+
marker.innerHTML = "<span></span>";
|
| 147 |
+
let dragged = false;
|
| 148 |
+
enableDrag(marker, layout, () => { dragged = true; });
|
| 149 |
+
marker.onclick = (clickEvent) => {
|
| 150 |
+
clickEvent.stopPropagation();
|
| 151 |
+
if (dragged) {
|
| 152 |
+
dragged = false;
|
| 153 |
+
return;
|
| 154 |
+
}
|
| 155 |
+
openPopover(item, marker, layout);
|
| 156 |
+
};
|
| 157 |
+
overlay.appendChild(marker);
|
| 158 |
+
}
|
| 159 |
+
|
| 160 |
function openPopover(item, marker, layout) {
|
| 161 |
closePopover();
|
| 162 |
marker.classList.add("active");
|
|
|
|
| 203 |
span.style.width = `${Math.max(8, (x2 - x1) * layout.scale)}px`;
|
| 204 |
span.style.height = `${Math.max(12, (y2 - y1) * layout.scale)}px`;
|
| 205 |
span.style.fontSize = `${clamp((y2 - y1) * layout.scale * 0.72, 8, 24)}px`;
|
| 206 |
+
span.addEventListener("click", (event) => triggerDotAtText(event, region, layout));
|
| 207 |
layer.appendChild(span);
|
| 208 |
});
|
| 209 |
const explainSelection = () => {
|
frontend/styles.css
CHANGED
|
@@ -50,7 +50,7 @@ main { width: 100%; max-width: 1320px; margin: auto; padding: 30px 14px calc(64p
|
|
| 50 |
.scanner-shell { display: flex; flex-direction: column; min-height: 0; overflow: hidden; background: var(--white); border: 1px solid rgba(20,35,31,.06); border-radius: 20px; box-shadow: var(--shadow); transition: grid-template-columns .42s cubic-bezier(.22,1,.36,1); }
|
| 51 |
.viewport { position: relative; min-height: min(62svh, 560px); background: radial-gradient(circle at center, #23423a, #071a18 72%); overflow: hidden; }
|
| 52 |
.viewport video { position: absolute; width: 100%; height: 100%; object-fit: contain; }
|
| 53 |
-
.document-scroller { position: absolute; inset: 0; z-index: 1; overflow: auto; overscroll-behavior: contain; scrollbar-gutter: stable; scrollbar-color: #71827c #162a25; scrollbar-width: auto; background: #152622; }
|
| 54 |
.document-scroller[hidden] { display: none; }
|
| 55 |
.document-scroller::-webkit-scrollbar { width: 11px; height: 11px; }
|
| 56 |
.document-scroller::-webkit-scrollbar-track { background: #162a25; }
|
|
@@ -80,7 +80,9 @@ main { width: 100%; max-width: 1320px; margin: auto; padding: 30px 14px calc(64p
|
|
| 80 |
.document-info span { display: grid; place-items: center; width: 25px; height: 25px; background: var(--lime); color: var(--dark); border-radius: 50%; font-family: Georgia, serif; font-weight: 800; }
|
| 81 |
.document-info b { display: none; font-size: 10px; }
|
| 82 |
.selection-layer { position: absolute; inset: 0; z-index: 2; pointer-events: none; }
|
| 83 |
-
.selectable-region { position: absolute; display: block; overflow: hidden; color: transparent; line-height: 1; white-space: pre-wrap; user-select: text; -webkit-user-select: text; pointer-events: auto; cursor: text; }
|
|
|
|
|
|
|
| 84 |
.selectable-region::selection { background: rgba(201,242,87,.72); color: rgba(8,28,24,.9); }
|
| 85 |
@keyframes pop { from { scale: .2; opacity: 0; } }
|
| 86 |
.corner { position: absolute; width: 34px; height: 34px; border-color: var(--lime); border-style: solid; opacity: .8; }
|
|
|
|
| 50 |
.scanner-shell { display: flex; flex-direction: column; min-height: 0; overflow: hidden; background: var(--white); border: 1px solid rgba(20,35,31,.06); border-radius: 20px; box-shadow: var(--shadow); transition: grid-template-columns .42s cubic-bezier(.22,1,.36,1); }
|
| 51 |
.viewport { position: relative; min-height: min(62svh, 560px); background: radial-gradient(circle at center, #23423a, #071a18 72%); overflow: hidden; }
|
| 52 |
.viewport video { position: absolute; width: 100%; height: 100%; object-fit: contain; }
|
| 53 |
+
.document-scroller { position: absolute; inset: 0; z-index: 1; overflow: auto; overscroll-behavior: contain; scrollbar-gutter: stable; scrollbar-color: #71827c #162a25; scrollbar-width: auto; background: #152622; -webkit-overflow-scrolling: touch; touch-action: pan-x pan-y; }
|
| 54 |
.document-scroller[hidden] { display: none; }
|
| 55 |
.document-scroller::-webkit-scrollbar { width: 11px; height: 11px; }
|
| 56 |
.document-scroller::-webkit-scrollbar-track { background: #162a25; }
|
|
|
|
| 80 |
.document-info span { display: grid; place-items: center; width: 25px; height: 25px; background: var(--lime); color: var(--dark); border-radius: 50%; font-family: Georgia, serif; font-weight: 800; }
|
| 81 |
.document-info b { display: none; font-size: 10px; }
|
| 82 |
.selection-layer { position: absolute; inset: 0; z-index: 2; pointer-events: none; }
|
| 83 |
+
.selectable-region { position: absolute; display: block; overflow: hidden; color: transparent; line-height: 1; white-space: pre-wrap; user-select: text; -webkit-user-select: text; pointer-events: auto; cursor: text; touch-action: pan-x pan-y; }
|
| 84 |
+
.term-marker.user-triggered { animation: marker-pulse .28s cubic-bezier(.22,1,.36,1); }
|
| 85 |
+
@keyframes marker-pulse { from { opacity: 0; transform: scale(.2); } 70% { transform: scale(1.25); } }
|
| 86 |
.selectable-region::selection { background: rgba(201,242,87,.72); color: rgba(8,28,24,.9); }
|
| 87 |
@keyframes pop { from { scale: .2; opacity: 0; } }
|
| 88 |
.corner { position: absolute; width: 34px; height: 34px; border-color: var(--lime); border-style: solid; opacity: .8; }
|
tests/test_services.py
CHANGED
|
@@ -125,3 +125,41 @@ def test_scanner_serves_scroll_info_and_clean_status_controls():
|
|
| 125 |
assert 'id="stabilityMetric"' not in html
|
| 126 |
assert 'id="lightMetric"' not in html
|
| 127 |
assert 'id="ocrMetric"' not in html
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 125 |
assert 'id="stabilityMetric"' not in html
|
| 126 |
assert 'id="lightMetric"' not in html
|
| 127 |
assert 'id="ocrMetric"' not in html
|
| 128 |
+
|
| 129 |
+
|
| 130 |
+
def test_multipage_pdf_is_rendered_as_one_scroll_surface():
|
| 131 |
+
import fitz
|
| 132 |
+
document = fitz.open()
|
| 133 |
+
first = document.new_page()
|
| 134 |
+
first.insert_text((72, 100), "Commercial Invoice and HS Code", fontsize=16)
|
| 135 |
+
second = document.new_page()
|
| 136 |
+
second.insert_text((72, 100), "Packing List and Customs Duty", fontsize=16)
|
| 137 |
+
response = client.post("/analyze-document", json={
|
| 138 |
+
"file_base64": base64.b64encode(document.tobytes()).decode(),
|
| 139 |
+
"filename": "two-pages.pdf",
|
| 140 |
+
"language_preference": "en",
|
| 141 |
+
})
|
| 142 |
+
assert response.status_code == 200
|
| 143 |
+
data = response.json()
|
| 144 |
+
assert data["page_count"] == 2
|
| 145 |
+
assert data["pages_analyzed"] == 2
|
| 146 |
+
assert data["frame_height"] > data["frame_width"] * 2
|
| 147 |
+
assert {item["term"] for item in data["detected_terms"]} >= {"HS Code", "Commercial Invoice"}
|
| 148 |
+
|
| 149 |
+
|
| 150 |
+
def test_document_cache_makes_repeat_upload_fast_path():
|
| 151 |
+
xml = '''<?xml version="1.0"?><w:document xmlns:w="http://schemas.openxmlformats.org/wordprocessingml/2006/main"><w:body><w:p><w:r><w:t>Delivery Order</w:t></w:r></w:p></w:body></w:document>'''
|
| 152 |
+
buffer = io.BytesIO()
|
| 153 |
+
with zipfile.ZipFile(buffer, "w") as archive:
|
| 154 |
+
archive.writestr("word/document.xml", xml)
|
| 155 |
+
payload = {"file_base64": base64.b64encode(buffer.getvalue()).decode(), "filename": "repeat.docx", "language_preference": "en"}
|
| 156 |
+
assert client.post("/analyze-document", json=payload).status_code == 200
|
| 157 |
+
second = client.post("/analyze-document", json=payload)
|
| 158 |
+
assert second.status_code == 200
|
| 159 |
+
assert second.json()["cached"] is True
|
| 160 |
+
|
| 161 |
+
|
| 162 |
+
def test_confidence_calibration_does_not_multiply_down_verified_match():
|
| 163 |
+
service = GlossaryService(ROOT / "data/glossary.json", ROOT / "data/user_corrections.json", ROOT / "data/sme_approved_definitions.json")
|
| 164 |
+
terms = service.match_regions([{"text": "HS Code", "bbox": [0, 0, 100, 20], "confidence": .8, "language": "en"}])
|
| 165 |
+
assert terms[0]["confidence"] > .8
|