qshub / workflow /step2_classify.py
basel00omar's picture
Fix UI styling for Dashboard start project, update QTO extraction pipeline logic, add PyMuPDF fallback for columns and OCR for floor heights
8682034
Raw History Blame
7.88 kB
"""
STEP 2 — Classify Pages
AI reads each page title/text and assigns drawing type
User can correct any misclassification
"""
import pandas as pd
import os, io, re, json
from pdf_engine.smart_classifier import classify_all_pages, PAGE_ITEMS_MAP
from pdf_engine.pdf_loader import page_to_pil
from workflow.workflow_state import mark_step_done
# Candidate types per PDF kind — used by the AI-vision refiner
_STRUCT_TYPES = ["foundations", "tie_beam", "column_layout", "upper_columns", "neck_columns",
"columns_1f", "columns_2f", "columns_roof",
"slab_1st", "slab_2nd", "roof_slab"]
_ARCH_TYPES = ["ground_floor_plan", "first_floor_plan", "second_floor_plan",
"roof_floor_plan", "elevations", "setting_out", "schedules"]
def _ai_classify(page_arr, pdf_type: str):
"""Classify ONE page by vision when keyword matching is unsure. Returns a
PAGE_ITEMS_MAP key or None."""
from utils.key_manager import get_key_manager
mgr = get_key_manager()
api_key, current_model = mgr.get_key_and_model()
if not api_key or api_key == "NO_API_KEY_FOUND":
return None
choices = _STRUCT_TYPES if pdf_type == "structural" else _ARCH_TYPES
try:
from google import genai
from google.genai import types
import time
import io as _io
pil_img = page_to_pil(page_arr)
buf = _io.BytesIO()
pil_img.save(buf, format="PNG")
img = buf.getvalue()
prompt = (
f"This is a construction drawing. Classify it as one of: {choices}. "
"Reply ONLY with valid JSON: {\"type\": \"<chosen_type>\"}"
)
for attempt in range(5):
api_key, current_model = mgr.get_key_and_model()
if not api_key or api_key == "NO_API_KEY_FOUND":
return None
try:
client = genai.Client(api_key=api_key, http_options={'timeout': 60})
resp = client.models.generate_content(
model=current_model,
contents=[types.Part.from_bytes(data=img, mime_type="image/png"), prompt])
raw = re.sub(r"```json|```", "", resp.text).strip()
result = json.loads(raw).get("type")
return result if result in PAGE_ITEMS_MAP else None
except Exception as e:
last_err = str(e)
if any(code in last_err for code in ("429", "RESOURCE_EXHAUSTED", "quota")):
mgr.mark_rate_limited(api_key, current_model)
continue
time.sleep(2)
return None
except Exception:
return None
from utils.i18n import t
def render_step2() -> bool:
import streamlit as st
st.markdown(t("classify_title"))
st.caption(t("classify_caption"))
str_pages = st.session_state.get("str_pages", [])
arch_pages = st.session_state.get("arch_pages", [])
str_texts = st.session_state.get("str_texts", [])
arch_texts = st.session_state.get("arch_texts", [])
# Auto-classify if not done yet
if "classified_pages" not in st.session_state:
with st.spinner(t("classify_running")):
classified = []
for i, txt in enumerate(str_texts):
classified.append({
"pdf": "structural",
"page_index": i,
"page_num": i + 1,
"text_preview": txt[:100].replace("\n", " "),
**_classify_single(txt, "structural"),
})
for i, txt in enumerate(arch_texts):
classified.append({
"pdf": "architectural",
"page_index": i,
"page_num": i + 1,
"text_preview": txt[:100].replace("\n", " "),
**_classify_single(txt, "architectural"),
})
# Verify EVERY page with AI Vision (more reliable than keywords on
# unseen drawings). Vision wins when it returns a valid type.
prog = st.progress(0.0, text="AI-verifying every page by image…")
for n, p in enumerate(classified):
src = str_pages if p["pdf"] == "structural" else arch_pages
if p["page_index"] < len(src):
g = _ai_classify(src[p["page_index"]], p["pdf"])
if g:
p["detected_type"] = g
p["confidence"] = "high(ai)"
p["items"] = PAGE_ITEMS_MAP.get(g, {}).get("extract_items", [])
prog.progress((n + 1) / len(classified))
prog.empty()
st.session_state["classified_pages"] = classified
classified = st.session_state["classified_pages"]
# Show classification results as editable table
st.markdown("### Auto-Classification Results | نتائج التصنيف التلقائي")
drawing_types = ["unknown"] + list(PAGE_ITEMS_MAP.keys())
for page in classified:
col_img, col_info, col_edit = st.columns([2, 2, 2])
pages_src = (
st.session_state.get("str_pages", []) if page["pdf"] == "structural"
else st.session_state.get("arch_pages", [])
)
with col_img:
if page["page_index"] < len(pages_src):
st.image(
page_to_pil(pages_src[page["page_index"]]),
caption=f"{page['pdf']} P{page['page_num']}",
use_container_width=True
)
with col_info:
_conf = str(page.get("confidence", "low"))
conf_icon = "" if _conf.startswith("high") else {"medium": "", "low": ""}.get(_conf, "⚪")
ai_tag = " (AI vision)" if "ai" in _conf else ""
st.markdown(f"**{conf_icon} Auto-detected{ai_tag}:**")
st.markdown(f"`{page.get('detected_type','unknown')}`")
st.caption(page["text_preview"])
with col_edit:
current_idx = drawing_types.index(page.get("detected_type","unknown")) if page.get("detected_type","unknown") in drawing_types else 0
new_type = st.selectbox(
"Drawing Type",
options=drawing_types,
index=current_idx,
key=f"cls_{page['pdf']}_{page['page_index']}",
format_func=lambda x: x.replace("_", " ").title() if x != "unknown" else " Unknown — select type"
)
page["detected_type"] = new_type
if new_type != "unknown":
page["items"] = PAGE_ITEMS_MAP.get(new_type, {}).get("extract_items", [])
st.divider()
# Update session state
st.session_state["classified_pages"] = classified
ready = [p for p in classified if p["detected_type"] != "unknown"]
unknown = [p for p in classified if p["detected_type"] == "unknown"]
if unknown:
st.warning(f" {len(unknown)} pages still unclassified — please assign types above")
st.info(f" {len(ready)} pages ready for extraction")
if ready and st.button(t("classify_next_btn"), type="primary", use_container_width=True):
mark_step_done("classify")
return True
return False
def _classify_single(text: str, pdf_type: str) -> dict:
from pdf_engine.smart_classifier import PAGE_ITEMS_MAP
text_lower = text.lower()
scores = {}
for ptype, config in PAGE_ITEMS_MAP.items():
score = sum(1 for kw in config.get("drawing_keywords", []) if kw.lower() in text_lower)
scores[ptype] = score
best = max(scores, key=scores.get)
score = scores[best]
return {
"detected_type": best if score > 0 else "unknown",
"confidence": "high" if score >= 2 else "medium" if score == 1 else "low",
"items": PAGE_ITEMS_MAP.get(best, {}).get("extract_items", []) if score > 0 else [],
}