first commit

This commit is contained in:
skylanix
2026-06-28 20:21:40 +02:00
commit b66b065da1
16 changed files with 1097 additions and 0 deletions
+5
View File
@@ -0,0 +1,5 @@
from .ocr import TweetOCRExtractor
from .classifier import ZeroShotClassifier
from .organizer import ImageThemeOrganizer
__version__ = "1.0.0"
+51
View File
@@ -0,0 +1,51 @@
import logging
from typing import List, Dict, Any, Optional
from transformers import pipeline
logger = logging.getLogger(__name__)
class ZeroShotClassifier:
"""
Classe responsable de la classification de texte à l'aide de modèles Hugging Face Zero-Shot.
"""
def __init__(self, model_name: str = "MoritzLaurer/mDeBERTa-v3-base-xnli-multilingual-nli-2mil7"):
"""
Initialise le classifieur zero-shot.
:param model_name: Nom du modèle Hugging Face à utiliser.
"""
self.model_name = model_name
self._pipeline = None
@property
def classifier_pipeline(self):
"""
Initialisation tardive (lazy loading) du pipeline pour économiser de la mémoire et du temps au démarrage.
"""
if self._pipeline is None:
logger.info(f"Chargement du pipeline de classification avec le modèle {self.model_name} (ceci peut prendre quelques secondes)...")
# On laisse Hugging Face gérer le choix du device (GPU s'il est dispo, sinon CPU)
self._pipeline = pipeline("zero-shot-classification", model=self.model_name)
return self._pipeline
def classify(self, text: str, candidate_labels: List[str] = None) -> Dict[str, Any]:
"""
Classifie un texte selon une liste de catégories candidates.
Si aucune catégorie n'est fournie, utilise les catégories de harcèlement par défaut.
:param text: Le texte à classifier.
:param candidate_labels: Liste des catégories (labels).
:return: Dictionnaire contenant les labels et leurs scores associés.
"""
if candidate_labels is None:
candidate_labels = ["Cyberharcèlement", "Insulte", "Menace", "Non-harcèlement"]
if not text or not text.strip():
# Si le texte est vide, on renvoie une structure vide ou par défaut
return {"labels": [], "scores": []}
try:
# On exécute le pipeline de classification
result = self.classifier_pipeline(text, candidate_labels=candidate_labels)
return result
except Exception as e:
logger.error(f"Erreur lors de la classification du texte : {e}")
raise RuntimeError(f"Échec de la classification : {e}") from e
+134
View File
@@ -0,0 +1,134 @@
import argparse
import sys
import logging
from pathlib import Path
from .organizer import ImageThemeOrganizer
from .web_generator import WebReportGenerator
def setup_logging(verbose: bool):
"""
Configure la journalisation.
"""
level = logging.DEBUG if verbose else logging.INFO
logging.basicConfig(
level=level,
format="%(asctime)s [%(levelname)s] %(name)s : %(message)s",
handlers=[
logging.StreamHandler(sys.stdout)
]
)
def main():
parser = argparse.ArgumentParser(
description="Classement automatique de captures d'écran de tweets par thèmes à l'aide d'EasyOCR et d'Hugging Face."
)
parser.add_argument(
"-i", "--input-dir",
type=str,
default=None,
help="Chemin vers le dossier contenant les captures d'écran à classer."
)
parser.add_argument(
"-o", "--output-dir",
type=str,
default=None,
help="Dossier de destination (par défaut, utilise ./ok/)."
)
parser.add_argument(
"--db",
type=str,
default=str(Path(__file__).parent.parent / "captures/ok/tweets.csv"),
help="Chemin vers le fichier base de données CSV (défaut: 'captures/ok/tweets.csv')."
)
parser.add_argument(
"--generate-report",
action="store_true",
help="Génère un rapport HTML à partir de la base de données CSV."
)
parser.add_argument(
"-m", "--model",
type=str,
default="MoritzLaurer/mDeBERTa-v3-base-xnli-multilingual-nli-2mil7",
help="Modèle Hugging Face Zero-Shot à utiliser."
)
parser.add_argument(
"-l", "--languages",
type=str,
default="fr,en",
help="Langues pour EasyOCR, séparées par des virgules (défaut: 'fr,en')."
)
parser.add_argument(
"-t", "--threshold",
type=float,
default=0.35,
help="Seuil de confiance de classification (0.0 à 1.0, défaut: 0.35)."
)
parser.add_argument(
"--copy",
action="store_true",
help="Copie les images au lieu de les déplacer."
)
parser.add_argument(
"--dry-run",
action="store_true",
help="Exécute une simulation sans modifier ni déplacer les fichiers."
)
parser.add_argument(
"-v", "--verbose",
action="store_true",
help="Active le mode verbeux pour afficher les logs de débogage."
)
args = parser.parse_args()
setup_logging(args.verbose)
if args.generate_report:
generator = WebReportGenerator(Path(args.db))
generator.generate()
return
# Normal processing path
if not args.input_dir:
print("Erreur : --input-dir est requis pour le traitement.", file=sys.stderr)
sys.exit(1)
input_path = Path(args.input_dir)
if not input_path.exists():
print(f"Erreur : Le dossier d'entrée '{args.input_dir}' n'existe pas.", file=sys.stderr)
sys.exit(1)
languages_list = [lang.strip() for lang in args.languages.split(",") if lang.strip()]
output_path = Path(args.output_dir) if args.output_dir else None
# Instanciation de l'organisateur
organizer = ImageThemeOrganizer(
input_dir=input_path,
output_dir=output_path,
ocr_languages=languages_list,
model_name=args.model,
confidence_threshold=args.threshold,
copy_only=args.copy,
dry_run=args.dry_run,
db_path=Path(args.db)
)
try:
organizer.run()
except Exception as e:
print(f"Une erreur critique est survenue : {e}", file=sys.stderr)
sys.exit(1)
if __name__ == "__main__":
main()
+62
View File
@@ -0,0 +1,62 @@
import csv
from pathlib import Path
from typing import List, Optional, Dict, Any
class CSVDatabaseManager:
def __init__(self, db_path: Path):
self.db_path = db_path
self.fieldnames = ['filename', 'filepath', 'status', 'ocr_text', 'detected_category', 'confidence', 'created_at']
self._initialize_csv()
def _initialize_csv(self):
if not self.db_path.exists():
self.db_path.parent.mkdir(parents=True, exist_ok=True)
with open(self.db_path, 'w', newline='', encoding='utf-8') as f:
writer = csv.DictWriter(f, fieldnames=self.fieldnames)
writer.writeheader()
def _read_all(self) -> List[Dict[str, Any]]:
with open(self.db_path, 'r', newline='', encoding='utf-8') as f:
return list(csv.DictReader(f))
def _write_all(self, data: List[Dict[str, Any]]):
with open(self.db_path, 'w', newline='', encoding='utf-8') as f:
writer = csv.DictWriter(f, fieldnames=self.fieldnames)
writer.writeheader()
writer.writerows(data)
def add_files(self, file_paths: List[Path]):
data = self._read_all()
existing_filenames = {row['filename'] for row in data}
new_entries = []
for path in file_paths:
if path.name not in existing_filenames:
new_entries.append({
'filename': path.name,
'filepath': str(path.absolute()),
'status': 'pending',
'ocr_text': '',
'detected_category': '',
'confidence': '',
'created_at': '' # Could add timestamp here
})
if new_entries:
data.extend(new_entries)
self._write_all(data)
def get_pending_files(self) -> List[Dict[str, Any]]:
return [row for row in self._read_all() if row['status'] == 'pending']
def update_file_status(self, filename: str, status: str, ocr_text: Optional[str] = None, category: Optional[str] = None, confidence: Optional[float] = None, new_filepath: Optional[str] = None):
data = self._read_all()
for row in data:
if row['filename'] == filename:
row['status'] = status
if ocr_text is not None: row['ocr_text'] = ocr_text
if category is not None: row['detected_category'] = category
if confidence is not None: row['confidence'] = str(confidence)
if new_filepath is not None: row['filepath'] = new_filepath
break
self._write_all(data)
+50
View File
@@ -0,0 +1,50 @@
import logging
from pathlib import Path
from typing import List, Optional
import easyocr
logger = logging.getLogger(__name__)
class TweetOCRExtractor:
"""
Classe responsable de l'extraction de texte à partir de captures d'écran en utilisant EasyOCR.
"""
def __init__(self, languages: Optional[List[str]] = None):
"""
Initialise le lecteur EasyOCR.
:param languages: Liste des langues à charger (par défaut ['fr', 'en']).
"""
if languages is None:
languages = ['fr', 'en']
self.languages = languages
self._reader = None
@property
def reader(self) -> easyocr.Reader:
"""
Initialisation tardive (lazy loading) d'EasyOCR pour économiser de la mémoire si non utilisé.
"""
if self._reader is None:
logger.info("Initialisation de l'OCR EasyOCR (ceci peut prendre quelques secondes)...")
self._reader = easyocr.Reader(self.languages)
return self._reader
def extract_text(self, image_path: Path) -> str:
"""
Extrait le texte d'une image.
:param image_path: Chemin vers le fichier image.
:return: Texte brut extrait.
"""
if not image_path.exists():
raise FileNotFoundError(f"Le fichier image n'existe pas : {image_path}")
try:
# easyocr accepte un chemin de fichier sous forme de string
results = self.reader.readtext(str(image_path))
# On joint les blocs de texte détectés
text_blocks = [text for (_, text, _) in results]
extracted_text = " ".join(text_blocks).strip()
return extracted_text
except Exception as e:
logger.error(f"Erreur lors de l'extraction OCR sur {image_path.name} : {e}")
raise RuntimeError(f"Échec de l'OCR : {e}") from e
+152
View File
@@ -0,0 +1,152 @@
import logging
import shutil
from pathlib import Path
from typing import List, Optional
from tqdm import tqdm
from .ocr import TweetOCRExtractor
from .classifier import ZeroShotClassifier
from .database_manager import CSVDatabaseManager
logger = logging.getLogger(__name__)
# Extensions d'images supportées
SUPPORTED_EXTENSIONS = {'.png', '.jpg', '.jpeg', '.webp', '.bmp', '.tiff'}
class ImageThemeOrganizer:
"""
Orchestrateur principal du traitement : parcourt les images, extrait le texte,
classifie, et organise les fichiers dans des sous-dossiers via une base de données.
"""
def __init__(
self,
input_dir: Path,
output_dir: Optional[Path] = None,
ocr_languages: Optional[List[str]] = None,
model_name: str = "MoritzLaurer/mDeBERTa-v3-base-xnli-multilingual-nli-2mil7",
confidence_threshold: float = 0.35,
copy_only: bool = False,
dry_run: bool = False,
db_path: Path = Path("captures/ok/tweets.csv")
):
self.input_dir = Path(input_dir)
self.output_dir = Path(output_dir) if output_dir else self.input_dir / "ok"
self.confidence_threshold = confidence_threshold
self.copy_only = copy_only
self.dry_run = dry_run
self.db = CSVDatabaseManager(db_path)
# Initialisation des modules (lazy)
self.ocr_extractor = TweetOCRExtractor(languages=ocr_languages)
self.classifier = ZeroShotClassifier(model_name=model_name)
def scan_new_files(self):
"""
Trouve tous les fichiers images dans le dossier d'entrée et les ajoute à la base de données.
"""
images = []
for file in self.input_dir.iterdir():
if file.is_file() and file.suffix.lower() in SUPPORTED_EXTENSIONS:
images.append(file)
if images:
self.db.add_files(images)
logger.info(f"Ajout de {len(images)} fichiers à la base de données.")
def process_pending(self):
"""
Traite tous les fichiers en attente dans la base de données.
"""
pending_files = self.db.get_pending_files()
if not pending_files:
logger.info("Aucune image en attente.")
return
for record in tqdm(pending_files, desc="Traitement des tweets"):
self._process_record(record)
def _process_record(self, record: dict):
image_path = Path(record['filepath'])
logger.info(f"Traitement de l'image : {image_path.name}")
try:
# 1. Extraction du texte
text = self.ocr_extractor.extract_text(image_path)
dest_category = "Non-classifié"
if not text:
logger.info(f" Aucun texte extrait pour {image_path.name}.")
dest_category = "Sans_Texte"
confidence = 1.0
else:
# 2. Classification du texte (Harassment categories)
classification_result = self.classifier.classify(text)
# Récupérer la meilleure catégorie
if classification_result and classification_result.get("labels"):
best_label = classification_result["labels"][0]
best_score = classification_result["scores"][0]
logger.info(f" Classification : {best_label} (score: {best_score:.2f})")
if best_score >= self.confidence_threshold:
dest_category = best_label
confidence = best_score
else:
dest_category = "Inclassable"
confidence = best_score
else:
dest_category = "Inclassable"
confidence = 0.0
# 3. Organisation physique du fichier
dest_path = self._organize_file(image_path, dest_category)
# 4. Mise à jour DB (on enregistre le nouvel emplacement du fichier
# afin que le rapport HTML pointe vers l'image déplacée).
self.db.update_file_status(
record['filename'], 'processed', text, dest_category, confidence,
new_filepath=str(dest_path)
)
except Exception as e:
logger.error(f" Échec du traitement de {image_path.name} : {e}")
self.db.update_file_status(record['filename'], 'error')
def _organize_file(self, image_path: Path, category: str) -> Path:
"""
Crée le dossier de destination et y déplace ou copie le fichier image.
Retourne le chemin de destination du fichier (utilisé pour la base de
données et le rapport HTML).
"""
dest_dir = self.output_dir / category
dest_path = dest_dir / image_path.name
if self.dry_run:
logger.info(f"[DRY-RUN] Déplacer {image_path.name} -> {dest_dir.name}/")
return dest_path
# Créer le dossier de catégorie si nécessaire
dest_dir.mkdir(parents=True, exist_ok=True)
try:
if self.copy_only:
shutil.copy2(image_path, dest_path)
else:
shutil.move(image_path, dest_path)
logger.info(f" Organisé : {image_path.name} -> {dest_dir.name}/")
except Exception as e:
logger.error(f"Impossible d'organiser le fichier {image_path.name} vers {dest_dir} : {e}")
return dest_path
def run(self) -> dict:
"""
Exécute le processus complet.
"""
self.scan_new_files()
self.process_pending()
return {}
+353
View File
@@ -0,0 +1,353 @@
import csv
from pathlib import Path
from datetime import datetime
from urllib.parse import quote
from collections import Counter
from jinja2 import Template
# Couleurs par catégorie (avec repli sur une couleur neutre).
CATEGORY_COLORS = {
"Cyberharcèlement": "#8e44ad",
"Menace": "#c0392b",
"Insulte": "#d35400",
"Harcèlement": "#e74c3c",
"Non-harcèlement": "#27ae60",
"Sans_Texte": "#7f8c8d",
"Inclassable": "#95a5a6",
"Non-classifié": "#bdc3c7",
}
DEFAULT_COLOR = "#34495e"
TEMPLATE = """
<!DOCTYPE html>
<html lang="fr">
<head>
<meta charset="UTF-8">
<meta name="viewport" content="width=device-width, initial-scale=1.0">
<title>Rapport de classification</title>
<style>
:root {
--bg: #f0f2f5;
--card-bg: #ffffff;
--text: #2c3e50;
--muted: #7f8c8d;
}
* { box-sizing: border-box; }
body {
font-family: 'Segoe UI', Tahoma, Geneva, Verdana, sans-serif;
margin: 0; padding: 0 20px 60px;
background-color: var(--bg); color: var(--text);
}
header {
position: sticky; top: 0; z-index: 50;
background: var(--bg); padding: 20px 0 12px;
border-bottom: 1px solid #dfe3e8;
}
h1 {
text-align: center; color: #1a2a6c;
margin: 0 0 4px; font-size: 1.8em;
}
.subtitle { text-align: center; color: var(--muted); margin: 0 0 16px; font-size: .9em; }
/* Barre de statistiques */
.stats {
display: flex; flex-wrap: wrap; gap: 10px;
justify-content: center; margin-bottom: 14px;
}
.stat {
background: var(--card-bg); border-radius: 10px; padding: 8px 16px;
box-shadow: 0 2px 6px rgba(0,0,0,.08); text-align: center; min-width: 90px;
}
.stat .num { font-size: 1.4em; font-weight: 700; }
.stat .lbl { font-size: .72em; color: var(--muted); text-transform: uppercase; letter-spacing: .5px; }
/* Contrôles : filtres, recherche, tri */
.controls {
display: flex; flex-wrap: wrap; gap: 10px;
align-items: center; justify-content: center;
}
.filters { display: flex; flex-wrap: wrap; gap: 8px; justify-content: center; }
.filter-btn {
border: none; cursor: pointer; color: #fff;
padding: 6px 14px; border-radius: 20px; font-size: .82em; font-weight: 600;
opacity: .55; transition: opacity .15s, transform .15s;
}
.filter-btn:hover { transform: translateY(-1px); }
.filter-btn.active { opacity: 1; box-shadow: 0 2px 8px rgba(0,0,0,.2); }
#search, #sort {
padding: 8px 12px; border: 1px solid #cfd6dd; border-radius: 8px;
font-size: .9em; background: #fff;
}
#search { min-width: 220px; }
/* Galerie */
.gallery {
display: grid; gap: 24px; margin-top: 24px;
grid-template-columns: repeat(auto-fill, minmax(330px, 1fr));
}
.card {
background: var(--card-bg); border-radius: 15px; overflow: hidden;
box-shadow: 0 8px 18px rgba(0,0,0,.08);
transition: transform .2s, box-shadow .2s;
display: flex; flex-direction: column;
border-top: 5px solid var(--cat-color, #ccc);
}
.card:hover { transform: translateY(-5px); box-shadow: 0 14px 28px rgba(0,0,0,.14); }
.img-container {
height: 240px; overflow: hidden; background: #1c2733;
display: flex; align-items: center; justify-content: center; cursor: zoom-in;
}
.img-container img { max-width: 100%; max-height: 100%; object-fit: contain; }
.card-body { padding: 16px 18px; display: flex; flex-direction: column; gap: 10px; flex-grow: 1; }
.card-head { display: flex; justify-content: space-between; align-items: center; gap: 8px; }
.category {
font-weight: 700; color: #fff; padding: 4px 12px;
border-radius: 20px; font-size: .8em; white-space: nowrap;
}
.filename { font-size: .78em; color: var(--muted); word-break: break-word; }
/* Barre de confiance */
.confidence { font-size: .78em; }
.confidence .bar-bg { background: #eceff1; border-radius: 6px; height: 8px; overflow: hidden; margin-top: 3px; }
.confidence .bar { height: 100%; border-radius: 6px; }
.conf-high { background: #27ae60; }
.conf-mid { background: #f39c12; }
.conf-low { background: #e74c3c; }
.ocr-text {
font-size: .82em; color: #444; background: #f7f9fa;
padding: 10px 12px; border-radius: 8px;
border-left: 4px solid var(--cat-color, #ccc);
max-height: 130px; overflow-y: auto; white-space: pre-wrap; line-height: 1.45;
}
.empty {
grid-column: 1 / -1; text-align: center; color: var(--muted);
padding: 60px 20px; font-size: 1.1em;
}
/* Lightbox */
#lightbox {
display: none; position: fixed; inset: 0; z-index: 100;
background: rgba(0,0,0,.88); align-items: center; justify-content: center;
cursor: zoom-out; padding: 30px;
}
#lightbox img { max-width: 95%; max-height: 95%; border-radius: 8px; box-shadow: 0 0 40px rgba(0,0,0,.6); }
</style>
</head>
<body>
<header>
<h1>Rapport de classification des tweets</h1>
<p class="subtitle">Généré le {{ generated_at }} — {{ items|length }} élément(s)</p>
<div class="stats">
<div class="stat"><div class="num">{{ items|length }}</div><div class="lbl">Total</div></div>
{% for cat, count in category_counts %}
<div class="stat">
<div class="num" style="color: {{ category_colors.get(cat, default_color) }}">{{ count }}</div>
<div class="lbl">{{ cat }}</div>
</div>
{% endfor %}
</div>
<div class="controls">
<div class="filters">
<button class="filter-btn active" data-cat="all" style="background:#34495e" onclick="filterCat(this,'all')">Tout</button>
{% for cat, count in category_counts %}
<button class="filter-btn active" data-cat="{{ cat }}"
style="background: {{ category_colors.get(cat, default_color) }}"
onclick="filterCat(this,'{{ cat }}')">{{ cat }} ({{ count }})</button>
{% endfor %}
</div>
<input id="search" type="text" placeholder="🔎 Rechercher dans le texte / fichier…" oninput="applyFilters()">
<select id="sort" onchange="sortCards()">
<option value="conf-desc">Confiance ↓</option>
<option value="conf-asc">Confiance ↑</option>
<option value="cat">Catégorie (A→Z)</option>
<option value="name">Nom de fichier</option>
</select>
</div>
</header>
<div class="gallery" id="gallery">
{% for item in items %}
<div class="card" data-category="{{ item.detected_category }}"
data-confidence="{{ item.confidence_value }}"
data-filename="{{ item.filename|lower }}"
data-text="{{ item.ocr_text|lower }}"
style="--cat-color: {{ category_colors.get(item.detected_category, default_color) }}">
<div class="img-container" onclick="openLightbox('{{ item.relative_filepath }}')">
<img src="{{ item.relative_filepath }}" alt="{{ item.filename }}" loading="lazy"
onerror="this.parentElement.innerHTML='<span style=&quot;color:#bbb;font-size:.85em&quot;>Image introuvable</span>'">
</div>
<div class="card-body">
<div class="card-head">
<span class="category" style="background: {{ category_colors.get(item.detected_category, default_color) }}">
{{ item.detected_category }}
</span>
<span class="filename">{{ item.filename }}</span>
</div>
<div class="confidence">
Confiance : <strong>{{ item.confidence_pct }}%</strong>
<div class="bar-bg">
<div class="bar {{ item.conf_class }}" style="width: {{ item.confidence_pct }}%"></div>
</div>
</div>
<div class="ocr-text">{{ item.ocr_text or "— Aucun texte extrait —" }}</div>
</div>
</div>
{% else %}
<div class="empty">Aucun élément traité à afficher.<br>Lancez d'abord le traitement des images.</div>
{% endfor %}
</div>
<div id="lightbox" onclick="this.style.display='none'"><img id="lightbox-img" src="" alt=""></div>
<script>
const activeCats = new Set(['all']);
function filterCat(btn, cat) {
const buttons = document.querySelectorAll('.filter-btn');
if (cat === 'all') {
activeCats.clear(); activeCats.add('all');
buttons.forEach(b => b.classList.toggle('active', b.dataset.cat === 'all'));
} else {
document.querySelector('.filter-btn[data-cat="all"]').classList.remove('active');
activeCats.delete('all');
btn.classList.toggle('active');
if (btn.classList.contains('active')) activeCats.add(cat); else activeCats.delete(cat);
if (activeCats.size === 0) {
activeCats.add('all');
document.querySelector('.filter-btn[data-cat="all"]').classList.add('active');
}
}
applyFilters();
}
function applyFilters() {
const q = document.getElementById('search').value.toLowerCase().trim();
document.querySelectorAll('.card').forEach(card => {
const catOk = activeCats.has('all') || activeCats.has(card.dataset.category);
const txtOk = !q || card.dataset.text.includes(q) || card.dataset.filename.includes(q);
card.style.display = (catOk && txtOk) ? '' : 'none';
});
}
function sortCards() {
const mode = document.getElementById('sort').value;
const gallery = document.getElementById('gallery');
const cards = Array.from(gallery.querySelectorAll('.card'));
cards.sort((a, b) => {
const ca = parseFloat(a.dataset.confidence) || 0, cb = parseFloat(b.dataset.confidence) || 0;
switch (mode) {
case 'conf-asc': return ca - cb;
case 'conf-desc': return cb - ca;
case 'cat': return a.dataset.category.localeCompare(b.dataset.category);
case 'name': return a.dataset.filename.localeCompare(b.dataset.filename);
}
});
cards.forEach(c => gallery.appendChild(c));
}
function openLightbox(src) {
document.getElementById('lightbox-img').src = src;
document.getElementById('lightbox').style.display = 'flex';
}
sortCards();
</script>
</body>
</html>
"""
class WebReportGenerator:
def __init__(self, csv_path: Path, output_dir: Path = Path("captures/ok")):
self.csv_path = Path(csv_path).resolve()
self.output_dir = Path(output_dir).resolve()
# Si le CSV n'existe pas, on le cherche dans output_dir.
if not self.csv_path.exists():
potential_path = self.output_dir / self.csv_path.name
if potential_path.exists():
self.csv_path = potential_path
def _resolve_relative_path(self, row: dict) -> str:
"""
Détermine le chemin de l'image relatif au rapport HTML (placé dans
output_dir), encodé pour une URL.
Robustesse : si le chemin enregistré n'existe pas (CSV obsolète d'avant
le déplacement), on reconstruit le chemin attendu
``output_dir/catégorie/fichier``.
"""
image_path = Path(row['filepath'])
category = row.get('detected_category') or ''
candidates = [image_path]
if category:
candidates.append(self.output_dir / category / image_path.name)
candidates.append(self.output_dir / image_path.name)
chosen = next((c for c in candidates if c.exists()), image_path)
try:
relative = chosen.relative_to(self.output_dir)
except ValueError:
# Repli : catégorie/fichier, sinon juste le nom du fichier.
relative = Path(category) / image_path.name if category else Path(image_path.name)
# Encodage URL (espaces, apostrophes typographiques, accents…) en
# préservant les séparateurs de dossiers.
return quote(relative.as_posix())
@staticmethod
def _confidence_fields(raw_value: str) -> dict:
try:
value = float(raw_value)
except (TypeError, ValueError):
value = 0.0
pct = round(value * 100)
if pct >= 60:
conf_class = "conf-high"
elif pct >= 35:
conf_class = "conf-mid"
else:
conf_class = "conf-low"
return {"confidence_value": value, "confidence_pct": pct, "conf_class": conf_class}
def generate(self):
items = []
if self.csv_path.exists():
with open(self.csv_path, 'r', encoding='utf-8') as f:
reader = csv.DictReader(f)
for row in reader:
if row.get('status') == 'processed':
row['relative_filepath'] = self._resolve_relative_path(row)
row.update(self._confidence_fields(row.get('confidence')))
items.append(row)
# Tri par défaut : confiance décroissante.
items.sort(key=lambda r: r['confidence_value'], reverse=True)
category_counts = Counter(item['detected_category'] for item in items)
# Catégories triées par effectif décroissant.
sorted_counts = sorted(category_counts.items(), key=lambda kv: (-kv[1], kv[0]))
self.output_dir.mkdir(parents=True, exist_ok=True)
timestamp = datetime.now().strftime("%Y%m%d_%H%M%S")
output_path = self.output_dir / f"report_{timestamp}.html"
template = Template(TEMPLATE)
html = template.render(
items=items,
category_counts=sorted_counts,
category_colors=CATEGORY_COLORS,
default_color=DEFAULT_COLOR,
generated_at=datetime.now().strftime("%d/%m/%Y à %H:%M"),
)
with open(output_path, 'w', encoding='utf-8') as f:
f.write(html)
print(f"Rapport généré : {output_path}")
return output_path