DLM_training_script 9.06 KB
Newer Older
Mayer's avatar
Mayer committed
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
import os
import gdown

from glob import glob

import cv2
from PIL import Image, ImageDraw, ImageFont
from IPython.display import display

from ultralytics import YOLO

import matplotlib.pyplot as plt
import matplotlib.patches as patches




# Download Dataset - Download and unpack a dataset consisting of a train, val and test split. Trees are labeled (using the YOLO format).
gdown.cached_download("https://drive.google.com/file/d/1xkh8RYp15c0N4A9HFvcBeSmxhWaH5Ynw/view?usp=sharing", "sorted_images_YOLO_formatted.zip", fuzzy=True, postprocess=gdown.extractall)






# Setting up the Framework - create a setup file named yolov8.yaml containing required parameters
%%writefile sorted_images_YOLO_formatted/data/yolov8.yaml

# Train/val/test sets
path: sorted_images_YOLO_formatted/data/ # dataset root dir
train: train # train images (relative to 'path')
val: val # val images (relative to 'path')
test: test # test images (optional)

# number of classes
nc: 2
# class names
names: ['Großer Laubbaum',
        'Kleiner Laubbaum']





# Training - Train the model on the train dataset. yolov8n.pt refers to the smallest model size
# Load a pretrained model
model = YOLO("yolov8n.pt")

# Train the model
results = model.train(data="sorted_images_YOLO_formatted/data/yolov8.yaml", epochs=500, imgsz=200, cache=False)  



# Model evaluation - Evaluate the model performance on the validation dataset
# Validate the model
metrics = model.val(split='val')
metrics.box.map    # map50-95
metrics.box.map50  # map50
metrics.box.map75  # map75
metrics.box.maps   # a list contains map50-95 of each category





# Display performance metrics
display(Image.open('runs/detect/train/results.png'))
display(Image.open('runs/detect/train2/confusion_matrix_normalized.png'))
display(Image.open('runs/detect/train2/F1_curve.png'))
display(Image.open('runs/detect/train2/labels_correlogram.jpg'))
display(Image.open('runs/detect/train2/labels.jpg'))
display(Image.open('runs/detect/train/P_curve.png'))
display(Image.open('runs/detect/train/PR_curve.png'))
display(Image.open('runs/detect/train/R_curve.png'))




# Model Inference
# We use the model for inference on a list of images.
# First, we load the model checkpoint with the best performance.

# Load trained model
model = YOLO('runs/detect/train2/weights/best.pt')  # load a custom trained model

# Export the model
#model.export(format='torchscript')
# model.export(format='onnx',simplify=True)

# We define a list of images on which we will apply the model to identify trees.
#imagefiles = [
#    "/content/ObjectDetectionDataset/data/test/513278_5404358_Base_A62_Luftbild_2021_EPSG25832.png",
#    "/content/ObjectDetectionDataset/data/test/513278_5404407_Base_A62_Luftbild_2021_EPSG25832.png",
#    "/content/ObjectDetectionDataset/data/test/513278_5404553_Base_A62_Luftbild_2021_EPSG25832.png"
#    ]

# Run batched inference on a list of images
#results = model.predict(imagefiles)


# Neue Methodik um alle Bilder aus /content/ObjectDetectionDataset/data/test vom Modell untersuchen zu lassen
# Verzeichnis mit den Testbildern
test_directory = "sorted_images_YOLO_formatted/data/test/"
imagefiles = glob(os.path.join(test_directory, "*.png"))

# Modellvorhersagen ausführen
results = model.predict(imagefiles)

# Benutzerdefinierte Farben für die Klassen (basierend auf meinem ursprünglichen Skript)
class_colors = {
    'Großer Laubbaum': 'red', 
    'Kleiner Laubbaum': 'orange'
}

# Klassen-IDs und Klassennamen zuordnen
class_names = ['Großer Laubbaum', 
               'Kleiner Laubbaum']


# Following 2 Options of how to display the output: 
# 1: Output with 200x200 pixels - save files to predictions/ directory - no readablie labels
from PIL import Image, ImageDraw, ImageFont
import os

# Ergebnisverzeichnis für die Vorhersagen erstellen, falls es nicht existiert
os.makedirs('predictions/', exist_ok=True)

# Benutzerdefinierte Farben für die Klassen
class_colors = {
    'Großer Laubbaum': 'red', 
    'Kleiner Laubbaum': 'orange'
}

# Vorhersagen plotten und Farben anpassen
for i, result in enumerate(results):
    # Bild laden
    img = Image.open(imagefiles[i])

    # Bild um 50 % größer skalieren
    img = img.resize((int(img.width * 1.5), int(img.height * 1.5)))

    draw = ImageDraw.Draw(img)

    # Bounding-Boxen und Klassen plotten
    for box in result.boxes:
        # Koordinaten der Bounding-Box
        xmin, ymin, xmax, ymax = box.xyxy[0].numpy()

        # Anpassung der Koordinaten an das skalierte Bild
        xmin, ymin, xmax, ymax = [coord * 1.5 for coord in [xmin, ymin, xmax, ymax]]

        # Klassennamen anhand der Klassen-ID holen
        class_id = int(box.cls)
        class_name = class_names[class_id]

        # Bounding-Box zeichnen mit der Farbe aus dem Farbschema
        draw.rectangle([xmin, ymin, xmax, ymax], outline=class_colors[class_name], width=3)

    # Bild ohne weißen Hintergrund und Achsen speichern
    img.save(os.path.join('predictions', os.path.split(imagefiles[i])[1]))

    print(f"Processed: {imagefiles[i]}")


# 2: Output with 200x200 pixels - save files to predictions/ directory - readable labels
from PIL import Image, ImageDraw, ImageFont
import os

# Ergebnisverzeichnis für die Vorhersagen erstellen, falls es nicht existiert
os.makedirs('predictions/', exist_ok=True)

# Benutzerdefinierte Farben für die Klassen
class_colors = {
    'Großer Laubbaum': 'red', 
    'Kleiner Laubbaum': 'orange', 
    'Großer Nadelbaum': 'darkgreen',
    'Kleiner Nadelbaum': 'lightgreen', 
    'Busch/Hecke (Laub/Hartlaub)': 'purple',
    'Busch/Hecke (Nadel)': 'cornflowerblue',
    'Unbekannt': 'white'
}

# Schriftart und -größe festlegen (falls verfügbar)
try:
    font = ImageFont.truetype("arial.ttf", 40)  # Verwende Arial mit Größe 20
except IOError:
    font = ImageFont.load_default()  # Fallback auf Standard-Schriftart

# Vorhersagen plotten und Farben anpassen
for i, result in enumerate(results):
    # Bild laden
    img = Image.open(imagefiles[i])

    # Bild um 50 % größer skalieren
    img = img.resize((int(img.width * 1.5), int(img.height * 1.5)))

    draw = ImageDraw.Draw(img)

    # Bounding-Boxen und Klassen plotten
    for box in result.boxes:
        # Koordinaten der Bounding-Box
        xmin, ymin, xmax, ymax = box.xyxy[0].numpy()

        # Anpassung der Koordinaten an das skalierte Bild
        xmin, ymin, xmax, ymax = [coord * 1.5 for coord in [xmin, ymin, xmax, ymax]]

        # Klassennamen anhand der Klassen-ID holen
        class_id = int(box.cls)
        class_name = class_names[class_id]

        # Bounding-Box zeichnen mit der Farbe aus dem Farbschema
        draw.rectangle([xmin, ymin, xmax, ymax], outline=class_colors[class_name], width=2)

        # Größe des Textes berechnen mit textbbox (ersetzt textsize)
        text_bbox = draw.textbbox((xmin, ymin), class_name, font=font)
        text_width = text_bbox[2] - text_bbox[0]
        text_height = text_bbox[3] - text_bbox[1]

        # Hintergrund für den Text (halbtransparent)
        text_background = (xmin, ymin - text_height, xmin + text_width, ymin)
        draw.rectangle(text_background, fill=(200, 200, 200, 128))  # Schwarzer halbtransparenter Hintergrund

        # Klassennamen über die Bounding-Box schreiben
        draw.text((xmin, ymin - text_height), class_name, font=font, fill="red")  # Weißer Text

    # Bild ohne weißen Hintergrund und Achsen speichern
    img.save(os.path.join('predictions', os.path.split(imagefiles[i])[1]))

    print(f"Processed: {imagefiles[i]}")






# Export of labels as .txt 
def yolo_format(class_id, xmin, ymin, xmax, ymax, img_width, img_height):
    x_center = (xmin + xmax) / 2.0 / img_width
    y_center = (ymin + ymax) / 2.0 / img_height
    width = (xmax - xmin) / img_width
    height = (ymax - ymin) / img_height
    return f"{class_id} {x_center} {y_center} {width} {height}"

def save_yolo_labels(results, output_dir):
    if not os.path.exists(output_dir):
        os.makedirs(output_dir)
    
    for i, result in enumerate(results):
        boxes = result.boxes  # Extract boxes from result
        img_width, img_height = result.orig_shape  # Get the original image shape

        image_path = result.path
        image_id = os.path.splitext(os.path.basename(image_path))[0]
        
        with open(os.path.join(output_dir, f"{image_id}.txt"), 'w') as f:
            for box in boxes:
                class_id = int(box.cls)
                xmin, ymin, xmax, ymax = box.xyxy[0].numpy()  # Extract box coordinates

                yolo_line = yolo_format(
                    class_id,
                    xmin,
                    ymin,
                    xmax,
                    ymax,
                    img_width,
                    img_height
                )
                f.write(yolo_line + '\n')

# Run batched inference on all images in the test directory
results = model.predict(imagefiles)

# Save the YOLO labels
output_dir = 'predictions'
save_yolo_labels(results, output_dir)


# Display results
#display all predictions
result_directory = "predictions"

# Get a list of all image files in the result directory
resultimages = glob(os.path.join(result_directory, "*.png"))

for image in resultimages: 
    display(Image.open(image))