Spaces:
Sleeping
Sleeping
Update app.py
Browse files
app.py
CHANGED
|
@@ -5,15 +5,16 @@ import re
|
|
| 5 |
import easyocr
|
| 6 |
import pdf2image
|
| 7 |
import numpy as np
|
| 8 |
-
import
|
| 9 |
-
from
|
|
|
|
| 10 |
|
| 11 |
# -----------------------------
|
| 12 |
-
# 1️⃣ OCR
|
| 13 |
# -----------------------------
|
| 14 |
-
reader = easyocr.Reader(['ar'
|
| 15 |
|
| 16 |
-
def
|
| 17 |
pages = pdf2image.convert_from_path(file_path, dpi=300)
|
| 18 |
text = ""
|
| 19 |
for page in pages:
|
|
@@ -22,51 +23,40 @@ def ocr_pdf_arabic(file_path):
|
|
| 22 |
return text
|
| 23 |
|
| 24 |
# -----------------------------
|
| 25 |
-
# 2️⃣
|
| 26 |
# -----------------------------
|
| 27 |
def calculer_emplacement(n):
|
| 28 |
try:
|
| 29 |
if n is None or n < 1:
|
| 30 |
return "### ⚠️ En attente d'un numéro de boîte..."
|
| 31 |
-
|
| 32 |
n = int(n)
|
| 33 |
etage = ((n - 1) // 11) + 1
|
| 34 |
rayon = "A" if n <= 11 else "B"
|
| 35 |
case = ((n - 1) % 11) + 1
|
| 36 |
-
|
| 37 |
return f"""
|
| 38 |
-
|
| 39 |
-
|
| 40 |
-
|
| 41 |
-
|
| 42 |
-
|
| 43 |
-
|
| 44 |
-
|
| 45 |
-
|
| 46 |
-
|
| 47 |
except:
|
| 48 |
return "❌ Erreur de saisie."
|
| 49 |
|
| 50 |
# -----------------------------
|
| 51 |
-
# 3️⃣
|
| 52 |
# -----------------------------
|
| 53 |
-
|
| 54 |
-
|
| 55 |
-
|
| 56 |
-
|
| 57 |
-
|
| 58 |
-
def summarize_arabic(text, longueur):
|
| 59 |
-
# Tokenizer et génération
|
| 60 |
-
tokenizer.src_lang = "ar_AR"
|
| 61 |
-
inputs = tokenizer(text, return_tensors="pt", max_length=1024, truncation=True)
|
| 62 |
-
max_len = 180 if longueur == "Détaillé" else 80
|
| 63 |
-
min_len = 50 if longueur == "Détaillé" else 20
|
| 64 |
-
summary_ids = model.generate(inputs["input_ids"], max_length=max_len, min_length=min_len, do_sample=False)
|
| 65 |
-
summary = tokenizer.decode(summary_ids[0], skip_special_tokens=True)
|
| 66 |
-
return summary
|
| 67 |
|
| 68 |
# -----------------------------
|
| 69 |
-
# 4️⃣ Classe
|
| 70 |
# -----------------------------
|
| 71 |
class ArchivAppV15:
|
| 72 |
def analyser_pdf(self, file, longueur):
|
|
@@ -80,15 +70,16 @@ class ArchivAppV15:
|
|
| 80 |
if content:
|
| 81 |
text += content + " "
|
| 82 |
|
| 83 |
-
# Si texte trop court → OCR
|
| 84 |
if len(text.strip()) < 50:
|
| 85 |
-
text =
|
| 86 |
|
| 87 |
clean_text = re.sub(r"\s+", " ", text).strip()
|
| 88 |
if len(clean_text) < 50:
|
| 89 |
return "❌ Document trop court après OCR", ""
|
| 90 |
|
| 91 |
-
|
|
|
|
| 92 |
|
| 93 |
return "✅ Synthèse réussie", summary
|
| 94 |
except Exception as e:
|
|
@@ -100,7 +91,7 @@ app = ArchivAppV15()
|
|
| 100 |
# 5️⃣ Interface Gradio
|
| 101 |
# -----------------------------
|
| 102 |
with gr.Blocks(theme=gr.themes.Soft()) as demo:
|
| 103 |
-
gr.Markdown("# 📁 ArchivChat V15 - Gestion & Synthèse Arabe")
|
| 104 |
|
| 105 |
with gr.Tab("📍 Localisation"):
|
| 106 |
gr.Markdown("### Calcul automatique : 11 boîtes par étage")
|
|
@@ -109,9 +100,9 @@ with gr.Blocks(theme=gr.themes.Soft()) as demo:
|
|
| 109 |
output_loc = gr.Markdown("### L'emplacement détaillé s'affichera ici")
|
| 110 |
input_num.change(calculer_emplacement, inputs=input_num, outputs=output_loc)
|
| 111 |
|
| 112 |
-
with gr.Tab("📄 Résumé
|
| 113 |
with gr.Row():
|
| 114 |
-
file_in = gr.File(label="Déposer le PDF d'archive (arabe)", file_types=[".pdf"])
|
| 115 |
longueur_opt = gr.Radio(["Court", "Détaillé"], label="Style de résumé", value="Court")
|
| 116 |
btn_res = gr.Button("Lancer l'analyse ✨", variant="primary")
|
| 117 |
with gr.Row():
|
|
|
|
| 5 |
import easyocr
|
| 6 |
import pdf2image
|
| 7 |
import numpy as np
|
| 8 |
+
from sumy.parsers.plaintext import PlaintextParser
|
| 9 |
+
from sumy.nlp.tokenizers import Tokenizer
|
| 10 |
+
from sumy.summarizers.lex_rank import LexRankSummarizer
|
| 11 |
|
| 12 |
# -----------------------------
|
| 13 |
+
# 1️⃣ OCR arabe + français
|
| 14 |
# -----------------------------
|
| 15 |
+
reader = easyocr.Reader(['ar', 'fr'])
|
| 16 |
|
| 17 |
+
def ocr_pdf_multilang(file_path):
|
| 18 |
pages = pdf2image.convert_from_path(file_path, dpi=300)
|
| 19 |
text = ""
|
| 20 |
for page in pages:
|
|
|
|
| 23 |
return text
|
| 24 |
|
| 25 |
# -----------------------------
|
| 26 |
+
# 2️⃣ Calcul emplacement (11 boîtes / étage)
|
| 27 |
# -----------------------------
|
| 28 |
def calculer_emplacement(n):
|
| 29 |
try:
|
| 30 |
if n is None or n < 1:
|
| 31 |
return "### ⚠️ En attente d'un numéro de boîte..."
|
|
|
|
| 32 |
n = int(n)
|
| 33 |
etage = ((n - 1) // 11) + 1
|
| 34 |
rayon = "A" if n <= 11 else "B"
|
| 35 |
case = ((n - 1) % 11) + 1
|
|
|
|
| 36 |
return f"""
|
| 37 |
+
# 📍 EMPLACEMENT TROUVÉ
|
| 38 |
+
---
|
| 39 |
+
## 📦 BOÎTE N° : {n}
|
| 40 |
+
## 🏢 RAYON : **{rayon}**
|
| 41 |
+
## 📏 ÉTAGE : **{etage}**
|
| 42 |
+
## 🔢 CASE (Position) : **{case} / 11**
|
| 43 |
+
---
|
| 44 |
+
*Instructions : Allez au Rayon {rayon}, montez à l'étage {etage} et prenez la {case}ème boîte.*
|
| 45 |
+
"""
|
| 46 |
except:
|
| 47 |
return "❌ Erreur de saisie."
|
| 48 |
|
| 49 |
# -----------------------------
|
| 50 |
+
# 3️⃣ Résumé extractif fiable hors-ligne
|
| 51 |
# -----------------------------
|
| 52 |
+
def summarize_offline(text, n_sentences=5):
|
| 53 |
+
parser = PlaintextParser.from_string(text, Tokenizer("arabic"))
|
| 54 |
+
summarizer = LexRankSummarizer()
|
| 55 |
+
summary = summarizer(parser.document, sentences_count=n_sentences)
|
| 56 |
+
return " ".join([str(sentence) for sentence in summary])
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 57 |
|
| 58 |
# -----------------------------
|
| 59 |
+
# 4️⃣ Classe analyse PDF
|
| 60 |
# -----------------------------
|
| 61 |
class ArchivAppV15:
|
| 62 |
def analyser_pdf(self, file, longueur):
|
|
|
|
| 70 |
if content:
|
| 71 |
text += content + " "
|
| 72 |
|
| 73 |
+
# Si texte trop court → OCR
|
| 74 |
if len(text.strip()) < 50:
|
| 75 |
+
text = ocr_pdf_multilang(file.name)
|
| 76 |
|
| 77 |
clean_text = re.sub(r"\s+", " ", text).strip()
|
| 78 |
if len(clean_text) < 50:
|
| 79 |
return "❌ Document trop court après OCR", ""
|
| 80 |
|
| 81 |
+
n_sentences = 10 if longueur == "Détaillé" else 5
|
| 82 |
+
summary = summarize_offline(clean_text, n_sentences=n_sentences)
|
| 83 |
|
| 84 |
return "✅ Synthèse réussie", summary
|
| 85 |
except Exception as e:
|
|
|
|
| 91 |
# 5️⃣ Interface Gradio
|
| 92 |
# -----------------------------
|
| 93 |
with gr.Blocks(theme=gr.themes.Soft()) as demo:
|
| 94 |
+
gr.Markdown("# 📁 ArchivChat V15 - Gestion & Synthèse Arabe + Français")
|
| 95 |
|
| 96 |
with gr.Tab("📍 Localisation"):
|
| 97 |
gr.Markdown("### Calcul automatique : 11 boîtes par étage")
|
|
|
|
| 100 |
output_loc = gr.Markdown("### L'emplacement détaillé s'affichera ici")
|
| 101 |
input_num.change(calculer_emplacement, inputs=input_num, outputs=output_loc)
|
| 102 |
|
| 103 |
+
with gr.Tab("📄 Résumé Multilingue"):
|
| 104 |
with gr.Row():
|
| 105 |
+
file_in = gr.File(label="Déposer le PDF d'archive (arabe ou français)", file_types=[".pdf"])
|
| 106 |
longueur_opt = gr.Radio(["Court", "Détaillé"], label="Style de résumé", value="Court")
|
| 107 |
btn_res = gr.Button("Lancer l'analyse ✨", variant="primary")
|
| 108 |
with gr.Row():
|