-
Notifications
You must be signed in to change notification settings - Fork 0
/
train.py
187 lines (160 loc) · 7.74 KB
/
train.py
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
import os
import argparse
import json
import tensorflow as tf
from tensorflow import keras
import numpy as np
import matplotlib.pyplot as plt
from datetime import datetime
from criterion import BasnetLoss
from pidnet import get_pred_model
from dataloder import KashmirDataset, CityscapesDataset
def parse_args():
parser = argparse.ArgumentParser(description="Train PIDNet on custom Dataset")
parser.add_argument('--config', type=str, required=True, help='Path to the configuration JSON file')
return parser.parse_args()
def load_config(config_path):
with open(config_path, 'r') as file:
config = json.load(file)
return config
def get_latest_experiment_dir(base_dir):
try:
experiment_dirs = [d for d in os.listdir(base_dir) if os.path.isdir(os.path.join(base_dir, d))]
if not experiment_dirs:
return None
experiment_dirs.sort(reverse=True)
return os.path.join(base_dir, experiment_dirs[0])
except:
print("Did you accidentally turn resume option on or wrong experiment directory name? 🤭🤭🤭 ")
print("Startig training from scratch again ..!")
def get_latest_checkpoint(checkpoint_dir):
if not os.path.exists(checkpoint_dir):
return None
checkpoints = [f for f in os.listdir(checkpoint_dir) if f.endswith('.h5')]
if not checkpoints:
return None
checkpoints.sort(reverse=True)
return os.path.join(checkpoint_dir, checkpoints[0])
def train_pidnet(config):
# Verify GPU availability
gpus = tf.config.experimental.list_physical_devices('GPU')
if gpus:
try:
# Set memory growth to avoid allocating all memory at once
for gpu in gpus:
tf.config.experimental.set_memory_growth(gpu, True)
print(f"🔥🔥 GPUs: {gpus} 🔥🔥")
except RuntimeError as e:
print(e)
base_experiment_dir = os.path.join("experiments", config['experiment_name'])
if config['resume']:
latest_experiment_dir = get_latest_experiment_dir(base_experiment_dir)
if latest_experiment_dir:
experiment_dir = latest_experiment_dir
print(f"Resuming from latest experiment directory: {experiment_dir}")
else:
current_time = datetime.now().strftime("%Y%m%d-%H%M%S")
experiment_dir = os.path.join(base_experiment_dir, current_time)
os.makedirs(experiment_dir, exist_ok=True)
print(f"No previous experiment found. Starting new experiment at: {experiment_dir}")
else:
current_time = datetime.now().strftime("%Y%m%d-%H%M%S")
experiment_dir = os.path.join(base_experiment_dir, current_time)
os.makedirs(experiment_dir, exist_ok=True)
print(f"Starting new experiment at: {experiment_dir}")
# save a copy of config back
with open(os.path.join(experiment_dir, 'config_copy.json'), 'w', encoding='utf-8') as f:
json.dump(config, f, ensure_ascii=False, indent=4)
if config['dataset'] == 'cityscapes':
train_ds = CityscapesDataset(config['root'], 'train', crop_size=(config['img_height'], config['img_width']), num_classes=config['num_classes'], mean=config['mean'], std=config['std'])
val_ds = CityscapesDataset(config['root'], 'val', crop_size=(config['img_height'], config['img_width']), num_classes=config['num_classes'], mean=config['mean'], std=config['std'])
else:
train_ds = KashmirDataset(config['root'], 'train', crop_size=(config['img_height'], config['img_width']), num_classes=config['num_classes'], mean=config['mean'], std=config['std'])
val_ds = KashmirDataset(config['root'], 'val', crop_size=(config['img_height'], config['img_width']), num_classes=config['num_classes'], mean=config['mean'], std=config['std'])
tf_train_ds = train_ds.get_dataset().batch(config['batch_size'])
tf_val_ds = val_ds.get_dataset().batch(config['batch_size'])
pidnet_model = get_pred_model(config['model_type'], (config['img_height'], config['img_width'], 3), config['num_classes'])
# Check for the latest checkpoint
checkpoint_dir = os.path.join(experiment_dir, 'checkpoints')
os.makedirs(checkpoint_dir, exist_ok=True)
latest_checkpoint = get_latest_checkpoint(checkpoint_dir)
initial_epoch = 0
if config['resume'] and latest_checkpoint:
print(f"🚀🚀 Resuming training from {latest_checkpoint} 🚀🚀🚀")
pidnet_model = keras.models.load_model(latest_checkpoint, custom_objects={"BasnetLoss": BasnetLoss, "MeanIoU": tf.keras.metrics.MeanIoU})
initial_epoch = int(latest_checkpoint.split('_')[-1].split('.')[0])
if config['show_model_summary']:
pidnet_model.summary() # Show model summary.
optimizer = keras.optimizers.Adam(learning_rate=config['learning_rate'], epsilon=config['epsilon'])
# Compile model
pidnet_model.compile(
loss=BasnetLoss(),
optimizer=optimizer,
metrics=[
tf.keras.metrics.MeanIoU(num_classes=config['num_classes'], name="mIoU", sparse_y_pred=False, sparse_y_true=True, axis=-1)
],
)
# TensorBoard callback
log_dir = os.path.join(experiment_dir, "logs")
tensorboard_callback = keras.callbacks.TensorBoard(log_dir=log_dir)
# Custom callback to save images
class SaveSamplesCallback(keras.callbacks.Callback):
def __init__(self, val_ds, output_dir, mean, std, num_samples=2):
super().__init__()
self.val_ds = val_ds
self.output_dir = output_dir
self.mean = np.array(mean)
self.std = np.array(std)
self.num_samples = num_samples
if not os.path.exists(output_dir):
os.makedirs(output_dir)
def on_epoch_end(self, epoch, logs=None):
for images, labels in self.val_ds.take(1):
predictions = self.model.predict(images)
predictions = tf.argmax(predictions, axis=-1)
predictions = tf.expand_dims(predictions, axis=-1)
images = (images * self.std + self.mean) * 255.0
images = np.clip(images, 0, 255).astype(np.uint8)
for i in range(min(self.num_samples, images.shape[0])):
fig, ax = plt.subplots(1, 2, figsize=(10, 5))
ax[0].imshow(images[i])
ax[0].set_title('Image')
ax[0].axis('off')
ax[1].imshow(predictions[i], cmap='gray')
ax[1].set_title('Prediction')
ax[1].axis('off')
plt.savefig(os.path.join(self.output_dir, f'epoch_{epoch + 1}_sample_{i + 1}.png'))
plt.close(fig)
# Callbacks
checkpoint_callback = keras.callbacks.ModelCheckpoint(
filepath=os.path.join(checkpoint_dir, 'model_checkpoint_{epoch:02d}.h5'),
save_freq='epoch',
save_best_only=True,
monitor='val_loss',
mode='min',
verbose=1,
save_weights_only=False,
)
reduce_lr_callback = keras.callbacks.ReduceLROnPlateau(
monitor='val_loss',
factor=0.5,
patience=10,
min_lr=1e-6,
verbose=1
)
save_samples_callback = SaveSamplesCallback(val_ds=tf_val_ds, output_dir=os.path.join(experiment_dir, 'sample_images'), mean=config['mean'], std=config['std'])
print("🚀🚀 Starting training now 🚀🚀🚀")
# Train the model
pidnet_model.fit(
tf_train_ds,
validation_data=tf_val_ds,
epochs=config['epochs'],
initial_epoch=initial_epoch,
callbacks=[checkpoint_callback, reduce_lr_callback, tensorboard_callback, save_samples_callback]
)
if __name__ == "__main__":
args = parse_args()
config = load_config(args.config)
train_pidnet(config)
# Example usage
# python train.py --config config.json