Python Tensorflow Bildsortierung
Hierfür braucht man relativ große Bilddatensätze als Beispiele für Tensorflow um zuverlässige Ergebnisse zu bekommen.
Situation
Ihr habt ein großen Ordner voller Bilder und mit diesen über https://teachablemachine.withgoogle.com ein Modell mit zwei Klassen erstellt und ALS QUANTITISIERT exportiert.
Script
python 3.9
pip install tensorflow pillow
#!/usr/bin/env python3
# sorter_tm.py
import os
import sys
import shutil
from PIL import Image
import numpy as np
import tensorflow as tf
tflite = tf.lite
def load_labels(label_path):
"""Liest die Klassennamen aus einer labels.txt im Format '0 Klasse1'."""
labels = []
with open(label_path, 'r', encoding='utf-8') as f:
for line in f:
line = line.strip()
if not line:
continue
parts = line.split(' ', 1)
if len(parts) < 2:
raise ValueError(f"Ungültiges Format in labels.txt: '{line}'")
labels.append(parts[1])
if not labels:
raise ValueError("Die labels.txt ist leer oder enthält keine gültigen Einträge.")
return labels
def prepare_interpreter(model_path):
"""Initialisiert den TFLite-Interpreter."""
interpreter = tflite.Interpreter(model_path=model_path)
interpreter.allocate_tensors()
input_details = interpreter.get_input_details()[0]
output_details = interpreter.get_output_details()[0]
return interpreter, input_details, output_details
def classify_image(interpreter, input_details, output_details, image, target_size):
input_shape = input_details['shape']
input_dtype = input_details['dtype']
# Resize des PIL-Bildes
img = image.resize(target_size)
arr = np.array(img)
if input_dtype == np.float32:
arr = arr.astype(np.float32) / 255.0
elif input_dtype == np.uint8:
arr = arr.astype(np.uint8)
else:
raise ValueError(f"Nicht unterstützter Eingabetyp: {input_dtype}")
arr = np.expand_dims(arr, axis=0)
interpreter.set_tensor(input_details['index'], arr)
interpreter.invoke()
output = interpreter.get_tensor(output_details['index'])[0]
output = output.astype(np.float32) # Konvertiere Ausgabe zu float32 für Softmax
probabilities = tf.nn.softmax(output).numpy()
idx = np.argmax(probabilities)
confidence = probabilities[idx]
return idx, confidence, probabilities
def main(model_path, label_path, src_dir, dst_root, image_size):
# Labels und Interpreter laden
labels = load_labels(label_path)
interpreter, in_det, out_det = prepare_interpreter(model_path)
# Ausgabeordnerstruktur anlegen
for lbl in labels:
os.makedirs(os.path.join(dst_root, lbl), exist_ok=True)
# Jedes Bild klassifizieren und verschieben
for filename in os.listdir(src_dir):
if not filename.lower().endswith((".jpg", ".jpeg", ".png", ".bmp", ".gif")):
continue
image_path = os.path.join(src_dir, filename)
try:
with Image.open(image_path) as img:
img = img.convert("RGB")
except Exception as e:
print(f"Bild konnte nicht geöffnet werden: {filename} – {e}")
continue
idx, confidence, probabilities = classify_image(interpreter, in_det, out_det, img, image_size)
label = labels[idx]
target_dir = os.path.join(dst_root, label)
os.makedirs(target_dir, exist_ok=True)
target_path = os.path.join(target_dir, filename)
try:
shutil.move(image_path, target_path)
print(f"{filename} → {label} ({confidence:.2%}) – Verteilung: {probabilities}")
except Exception as e:
print(f"Fehler beim Verschieben von {filename}: {e}")
if __name__ == "__main__":
if len(sys.argv) != 6:
print("Usage: python sorter_tm.py <model.tflite> <labels.txt> "
"<source_dir> <destination_root> <image_size>")
print("Beispiel: python sorter_tm.py model/model.tflite model/labels.txt "
"input_images sorted_images 224")
sys.exit(1)
model_file = sys.argv[1]
labels_file = sys.argv[2]
source_folder = sys.argv[3]
destination_root = sys.argv[4]
size = int(sys.argv[5])
main(model_file, labels_file, source_folder, destination_root, (size, size))
Vorgängerversionen
V2: Kann noch nicht mit Tensorflow Labels umgehen und erwartet feste Klassen.
#!/usr/bin/env python3
# sorter_tm.py
import os
import sys
import shutil
from PIL import Image
import numpy as np
import tensorflow as tf
tflite = tf.lite
def load_labels(label_path):
"""Liest die Klassennamen aus einer labels.txt."""
with open(label_path, 'r', encoding='utf-8') as f:
labels = [line.strip() for line in f if line.strip()]
if len(labels) != 2 or set(labels) != {"Good", "Bad"}:
raise ValueError("labels.txt muss genau die Einträge 'Good' und 'Bad' enthalten.")
return labels
def prepare_interpreter(model_path):
"""Initialisiert den TFLite-Interpreter."""
interpreter = tflite.Interpreter(model_path=model_path)
interpreter.allocate_tensors()
input_details = interpreter.get_input_details()[0]
output_details = interpreter.get_output_details()[0]
return interpreter, input_details, output_details
def classify_image(interpreter, input_details, output_details, image, target_size):
input_shape = input_details['shape']
input_dtype = input_details['dtype']
# Resize des PIL-Bildes
img = image.resize(target_size)
arr = np.array(img)
if input_dtype == np.float32:
arr = arr.astype(np.float32) / 255.0
elif input_dtype == np.uint8:
arr = arr.astype(np.uint8)
else:
raise ValueError(f"Nicht unterstützter Eingabetyp: {input_dtype}")
arr = np.expand_dims(arr, axis=0)
interpreter.set_tensor(input_details['index'], arr)
interpreter.invoke()
output = interpreter.get_tensor(output_details['index'])[0]
output = output.astype(np.float32) # Konvertiere Ausgabe zu float32 für Softmax
probabilities = tf.nn.softmax(output).numpy()
idx = np.argmax(probabilities)
confidence = probabilities[idx]
return idx, confidence, probabilities
def main(model_path, label_path, src_dir, dst_root, image_size):
# Labels und Interpreter laden
labels = load_labels(label_path)
interpreter, in_det, out_det = prepare_interpreter(model_path)
# Ausgabeordnerstruktur anlegen
for lbl in labels:
os.makedirs(os.path.join(dst_root, lbl), exist_ok=True)
# Jedes Bild klassifizieren und verschieben
for filename in os.listdir(src_dir):
if not filename.lower().endswith((".jpg", ".jpeg", ".png", ".bmp", ".gif")):
continue
image_path = os.path.join(src_dir, filename)
try:
with Image.open(image_path) as img:
img = img.convert("RGB")
except Exception as e:
print(f"Bild konnte nicht geöffnet werden: {filename} – {e}")
continue
idx, confidence, probabilities = classify_image(interpreter, in_det, out_det, img, image_size)
label = labels[idx]
target_dir = os.path.join(dst_root, label)
os.makedirs(target_dir, exist_ok=True)
target_path = os.path.join(target_dir, filename)
try:
shutil.move(image_path, target_path)
print(f"{filename} → {label} ({confidence:.2%}) – Verteilung: {probabilities}")
except Exception as e:
print(f"Fehler beim Verschieben von {filename}: {e}")
if __name__ == "__main__":
if len(sys.argv) != 6:
print("Usage: python sorter_tm.py <model.tflite> <labels.txt> "
"<source_dir> <destination_root> <image_size>")
print("Beispiel: python sorter_tm.py model/model.tflite model/labels.txt "
"input_images sorted_images 224")
sys.exit(1)
model_file = sys.argv[1]
labels_file = sys.argv[2]
source_folder = sys.argv[3]
destination_root = sys.argv[4]
size = int(sys.argv[5])
main(model_file, labels_file, source_folder, destination_root, (size, size))Notige Struktur:
C:\BildSorter\
├── model\
│ ├── model.tflite ← exportiertes Modell von Teachable Machine
│ └── labels.txt ← labels
├── input_images\ ← HIER liegen Ihre zu sortierenden Bilder
├── sorted_images\ ← WIRD automatisch befüllt mit Unterordnern
├── sorter_tm.py ← das Python-Skript
Run:
python sorter_tm.py model/model.tflite model/labels.txt input_images sorted_images 224