Потеря нана в керасе при работе над моделью семантической сегментации для сегментации опухолей головного мозга, задача 2

Потеря нана в керасе при работе над моделью семантической сегментации для сегментации опухолей головного мозга, задача 2 ⇐ Python

1 сообщение • Страница 1 из 1

Anonymous

Потеря нана в керасе при работе над моделью семантической сегментации для сегментации опухолей головного мозга, задача 2

Цитата

Сообщение Anonymous » 25 июн 2024, 11:16

Некоторый контекст:
На входе есть 4 изображения, и я объединил их в 4 канала, а на выходе — 1 изображение.
Входная форма (образец,240,240,4)
Выходная форма : (sample,240,240,1)
Потеря моей модели за 50 эпох — это Нэн, и точность тоже
код:
предварительная обработка и структурирование модели

Код: Выделить всё

import tensorflow.keras as kr
import numpy as np
import matplotlib.pylab as plt
#load data
inputimg = np.load('input_dataset.npy')
outputimg = np.load('output_dataset.npy')

inputimg = np.array(inputimg, dtype=np.float32)
outputimg = np.array(outputimg, dtype=np.float32)
outputimg = np.expand_dims(outputimg, axis=3)

x_train = inputimg[:1126].copy()
y_train = outputimg[:1126].copy()
x_train.shape[0] == y_train.shape[0]
x_eval = inputimg[1126:1189].copy()
y_eval = outputimg[1126:1189].copy()
x_test = inputimg[1189:].copy()
y_test = outputimg[1189:].copy()

print("train target:",y_train.shape,"                test target:",y_test.shape)
print(f"train x{x_test.shape}","                 test x:", x_test.shape)

#shuffle data
np.random.seed(0)
np.random.shuffle(x_train)
np.random.shuffle(y_train)
np.random.shuffle(x_eval)
np.random.shuffle(y_eval)

np.random.shuffle(x_test)
np.random.shuffle(y_test)

from tensorflow.keras import layers
inputs = kr.Input((240,240,4))
x = layers.Conv2D(64, 3, strides=2, activation="relu", padding="same")(inputs)
x = layers.Conv2D(64, 3, activation="relu", padding="same")(x)
x = layers.Conv2D(128, 3, strides=2, activation="relu", padding="same")(x)
x = layers.Conv2D(128, 3, activation="relu", padding="same")(x)
x = layers.Conv2D(256, 3, strides=2, padding="same", activation="relu")(x)
x = layers.Conv2D(256, 3, activation="relu", padding="same")(x)

x = layers.Conv2DTranspose(256, 3, activation="relu", padding="same")(x)
x = layers.Conv2DTranspose(256, 3, activation="relu", padding="same", strides=2)(x)
x = layers.Conv2DTranspose(128, 3, activation="relu", padding="same")(x)
x = layers.Conv2DTranspose(128, 3, activation="relu", padding="same", strides=2)(x)
x = layers.Conv2DTranspose(64, 3, activation="relu", padding="same")(x)
x = layers.Conv2DTranspose(64, 3, activation="relu", padding="same", strides=2)(x)

outputs = layers.Conv2D(4, 3, activation="softmax", padding="same")(x)
model = kr.Model(inputs,outputs)
model.summary()

#the BRATS 2021 task 1 brain tumour segmentation dataset
#https://www.kaggle.com/datasets/dschettler8845/brats-2021-task1/code?datasetId=1541666&sortBy=voteCount

выходные данные

Код: Выделить всё

train target: (1126, 240, 240, 1)                 test target: (62, 240, 240, 1)
train x(62, 240, 240, 4)                  test x: (62, 240, 240, 4)
Model:  "functional_3"
┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━┓
┃ Layer (type)                    ┃ Output Shape           ┃       Param # ┃
┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━┩
│ input_layer_1 (InputLayer)      │ (None, 240, 240, 4)    │             0 │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ conv2d_7 (Conv2D)               │ (None, 120, 120, 64)   │         2,368 │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ conv2d_8 (Conv2D)               │ (None, 120, 120, 64)   │        36,928 │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ conv2d_9 (Conv2D)               │ (None, 60, 60, 128)    │        73,856 │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ conv2d_10 (Conv2D)              │ (None, 60, 60, 128)    │       147,584 │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ conv2d_11 (Conv2D)              │ (None, 30, 30, 256)    │       295,168 │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ conv2d_12 (Conv2D)              │ (None, 30, 30, 256)    │       590,080 │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ conv2d_transpose_6              │ (None, 30, 30, 256)    │       590,080 │
│ (Conv2DTranspose)               │                        │               │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ conv2d_transpose_7              │ (None, 60, 60, 256)    │       590,080 │
│ (Conv2DTranspose)               │                        │               │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ conv2d_transpose_8              │ (None, 60, 60, 128)    │       295,040 │
│ (Conv2DTranspose)               │                        │               │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ conv2d_transpose_9              │ (None, 120, 120, 128)  │       147,584 │
│ (Conv2DTranspose)               │                        │               │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ conv2d_transpose_10             │ (None, 120, 120, 64)   │        73,792 │
│ (Conv2DTranspose)               │                        │               │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ conv2d_transpose_11             │ (None, 240, 240, 64)   │        36,928 │
│ (Conv2DTranspose)               │                        │               │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ conv2d_13 (Conv2D)              │ (None, 240, 240, 4)    │         2,308 │
└─────────────────────────────────┴────────────────────────┴───────────────┘
Total params: 2,881,796 (10.99 MB)
Trainable params: 2,881,796 (10.99 MB)
Non-trainable params: 0 (0.00 B)

при компиляции и обучении:

Код: Выделить всё

model.compile(optimizer="Adam", loss="sparse_categorical_crossentropy")

callbacks = [
kr.callbacks.ModelCheckpoint("help.keras",
save_best_only=True)
]

history = model.fit(inputimg,outputimg,
epochs=50,
callbacks=callbacks,)

результат компиляции и обучения

Код: Выделить всё

Epoch 1/50
WARNING: All log messages before absl::InitializeLog() is called are written to STDERR
I0000 00:00:1719246111.166733  116651 service.cc:145] XLA service 0x7f4470003550 initialized for platform CUDA (this does not guarantee that XLA will be used).  Devices:
I0000 00:00:1719246111.169110  116651 service.cc:153]   StreamExecutor device (0): NVIDIA GeForce RTX 4060 Laptop GPU, Compute Capability 8.9
2024-06-24 23:21:51.376397: I tensorflow/compiler/mlir/tensorflow/utils/dump_mlir_util.cc:268] disabling MLIR crash reproducer, set env var `MLIR_CRASH_REPRODUCER_DIRECTORY` to enable.
W0000 00:00:1719246111.595718  116651 assert_op.cc:38] Ignoring Assert operator compile_loss/sparse_categorical_crossentropy/SparseSoftmaxCrossEntropyWithLogits/assert_equal_1/Assert/Assert
2024-06-24 23:21:51.959587: I external/local_xla/xla/stream_executor/cuda/cuda_dnn.cc:465] Loaded cuDNN version 8907
2024-06-24 23:22:05.935224: E external/local_xla/xla/service/slow_operation_alarm.cc:65] Trying algorithm eng0{} for conv (f32[32,128,121,121]{3,2,1,0}, u8[0]{0}) custom-call(f32[32,128,60,60]{3,2,1,0}, f32[128,128,3,3]{3,2,1,0}), window={size=3x3 stride=2x2}, dim_labels=bf01_oi01->bf01, custom_call_target="__cudnn$convBackwardInput", backend_config={"operation_queue_id":"0","wait_on_operation_queues":[],"cudnn_conv_backend_config":{"conv_result_scale":1,"activation_mode":"kNone","side_input_scale":0,"leakyrelu_alpha":0}} is taking a while...
2024-06-24 23:22:06.654590: E external/local_xla/xla/service/slow_operation_alarm.cc:133] The operation took 1.718804966s
Trying algorithm eng0{} for conv (f32[32,128,121,121]{3,2,1,0}, u8[0]{0}) custom-call(f32[32,128,60,60]{3,2,1,0}, f32[128,128,3,3]{3,2,1,0}), window={size=3x3 stride=2x2}, dim_labels=bf01_oi01->bf01, custom_call_target="__cudnn$convBackwardInput", backend_config={"operation_queue_id":"0","wait_on_operation_queues":[],"cudnn_conv_backend_config":{"conv_result_scale":1,"activation_mode":"kNone","side_input_scale":0,"leakyrelu_alpha":0}} is taking a while...
2024-06-24 23:22:12.322431: E external/local_xla/xla/service/slow_operation_alarm.cc:65] Trying algorithm eng0{} for conv (f32[32,64,241,241]{3,2,1,0}, u8[0]{0}) custom-call(f32[32,64,120,120]{3,2,1,0}, f32[64,64,3,3]{3,2,1,0}), window={size=3x3 stride=2x2}, dim_labels=bf01_oi01->bf01, custom_call_target="__cudnn$convBackwardInput", backend_config={"operation_queue_id":"0","wait_on_operation_queues":[],"cudnn_conv_backend_config":{"conv_result_scale":1,"activation_mode":"kNone","side_input_scale":0,"leakyrelu_alpha":0}} is taking a while...
2024-06-24 23:22:12.953718: E external/local_xla/xla/service/slow_operation_alarm.cc:133] The operation took 1.631474573s
Trying algorithm eng0{} for conv (f32[32,64,241,241]{3,2,1,0}, u8[0]{0}) custom-call(f32[32,64,120,120]{3,2,1,0}, f32[64,64,3,3]{3,2,1,0}), window={size=3x3 stride=2x2}, dim_labels=bf01_oi01->bf01, custom_call_target="__cudnn$convBackwardInput", backend_config={"operation_queue_id":"0","wait_on_operation_queues":[],"cudnn_conv_backend_config":{"conv_result_scale":1,"activation_mode":"kNone","side_input_scale":0,"leakyrelu_alpha":0}} is taking a while...
2024-06-24 23:22:33.411114: E external/local_xla/xla/service/slow_operation_alarm.cc:65] Trying algorithm eng0{} for conv (f32[32,256,60,60]{3,2,1,0}, u8[0]{0}) custom-call(f32[32,128,60,60]{3,2,1,0}, f32[256,128,3,3]{3,2,1,0}), window={size=3x3 pad=1_1x1_1}, dim_labels=bf01_oi01->bf01, custom_call_target="__cudnn$convForward", backend_config={"operation_queue_id":"0","wait_on_operation_queues":[],"cudnn_conv_backend_config":{"conv_result_scale":1,"activation_mode":"kNone","side_input_scale":0,"leakyrelu_alpha":0}} is taking a while...
2024-06-24 23:22:34.596435: E external/local_xla/xla/service/slow_operation_alarm.cc:133] The operation took 2.184202328s
Trying algorithm eng0{} for conv (f32[32,256,60,60]{3,2,1,0}, u8[0]{0}) custom-call(f32[32,128,60,60]{3,2,1,0}, f32[256,128,3,3]{3,2,1,0}), window={size=3x3 pad=1_1x1_1}, dim_labels=bf01_oi01->bf01, custom_call_target="__cudnn$convForward", backend_config={"operation_queue_id":"0","wait_on_operation_queues":[],"cudnn_conv_backend_config":{"conv_result_scale":1,"activation_mode":"kNone","side_input_scale":0,"leakyrelu_alpha":0}} is taking a while...
I0000 00:00:1719246187.497218  116651 device_compiler.h:188] Compiled cluster using XLA!  This line is logged at most once for the lifetime of the process.
39/40 ━━━━━━━━━━━━━━━━━━━━ 0s 370ms/step - loss:  nan
W0000 00:00:1719246202.227021  116653 assert_op.cc:38] Ignoring Assert operator compile_loss/sparse_categorical_crossentropy/SparseSoftmaxCrossEntropyWithLogits/assert_equal_1/Assert/Assert
40/40 ━━━━━━━━━━━━━━━━━━━━ 114s 717ms/step - loss: nan
Epoch 2/50
WARNING: All log messages before absl::InitializeLog() is called are written to STDERR
I0000 00:00:1719246215.883065  116653 asm_compiler.cc:369] ptxas warning : Registers are spilled to local memory in function 'input_reduce_select_fusion_7', 188 bytes spill stores, 188 bytes spill loads
ptxas warning : Registers are spilled to local memory in function 'input_compare_reduce_fusion_1', 228 bytes spill stores, 228 bytes spill loads
ptxas warning : Registers are spilled to local memory in function 'input_reduce_select_fusion_9', 200 bytes spill stores, 200 bytes spill loads

/home/red/.pyenv/versions/3.11.4/lib/python3.11/site-packages/keras/src/callbacks/model_checkpoint.py:206: UserWarning: Can save best model only with val_loss available, skipping.
self._save_model(epoch=epoch, batch=None, logs=logs)
40/40 ━━━━━━━━━━━━━━━━━━━━ 14s 342ms/step - loss: nan
Epoch 3/50
40/40 ━━━━━━━━━━━━━━━━━━━━ 14s 359ms/step - loss: nan
Epoch 4/50
40/40 ━━━━━━━━━━━━━━━━━━━━ 15s 368ms/step - loss: nan
Epoch 5/50
40/40 ━━━━━━━━━━━━━━━━━━━━ 15s 361ms/step - loss: nan
Epoch 6/50
40/40 ━━━━━━━━━━━━━━━━━━━━ 15s 361ms/step - loss: nan
Epoch 7/50
40/40 ━━━━━━━━━━━━━━━━━━━━ 14s 358ms/step - loss: nan
Epoch 8/50
40/40 ━━━━━━━━━━━━━━━━━━━━ 15s 362ms/step - loss: nan
Epoch 9/50
40/40 ━━━━━━━━━━━━━━━━━━━━ 15s 378ms/step - loss: nan
Epoch 10/50
40/40 ━━━━━━━━━━━━━━━━━━━━ 16s 390ms/step - loss: nan
Epoch 11/50
40/40 ━━━━━━━━━━━━━━━━━━━━ 16s 397ms/step - loss: nan
Epoch 12/50
40/40 ━━━━━━━━━━━━━━━━━━━━ 15s 361ms/step - loss: nan
Epoch 13/50
40/40 ━━━━━━━━━━━━━━━━━━━━ 15s 365ms/step - loss: nan
Epoch 14/50
40/40 ━━━━━━━━━━━━━━━━━━━━ 14s 359ms/step - loss: nan
Epoch 15/50
40/40 ━━━━━━━━━━━━━━━━━━━━ 14s 348ms/step - loss: nan
Epoch 16/50
40/40 ━━━━━━━━━━━━━━━━━━━━ 14s 355ms/step - loss: nan
Epoch 17/50
40/40 ━━━━━━━━━━━━━━━━━━━━ 14s 355ms/step - loss: nan
Epoch 18/50
40/40 ━━━━━━━━━━━━━━━━━━━━ 15s 372ms/step - loss: nan
Epoch 19/50
40/40 ━━━━━━━━━━━━━━━━━━━━ 15s 362ms/step - loss: nan
Epoch 20/50
40/40 ━━━━━━━━━━━━━━━━━━━━ 14s 344ms/step - loss: nan
Epoch 21/50
40/40 ━━━━━━━━━━━━━━━━━━━━ 14s 339ms/step - loss: nan
Epoch 22/50
40/40 ━━━━━━━━━━━━━━━━━━━━ 14s 343ms/step - loss: nan
Epoch 23/50
40/40 ━━━━━━━━━━━━━━━━━━━━ 14s 341ms/step - loss: nan
Epoch 24/50
40/40 ━━━━━━━━━━━━━━━━━━━━ 16s 391ms/step - loss: nan
Epoch 25/50
40/40 ━━━━━━━━━━━━━━━━━━━━ 15s 376ms/step - loss: nan
Epoch 26/50
40/40 ━━━━━━━━━━━━━━━━━━━━ 16s 392ms/step - loss: nan
Epoch 27/50
40/40 ━━━━━━━━━━━━━━━━━━━━ 14s 360ms/step - loss: nan
Epoch 28/50
40/40 ━━━━━━━━━━━━━━━━━━━━ 15s 378ms/step - loss: nan
Epoch 29/50
40/40 ━━━━━━━━━━━━━━━━━━━━ 13s 332ms/step - loss: nan
Epoch 30/50
40/40 ━━━━━━━━━━━━━━━━━━━━ 14s 355ms/step - loss: nan
Epoch 31/50
40/40 ━━━━━━━━━━━━━━━━━━━━ 15s 381ms/step - loss: nan
Epoch 32/50
40/40 ━━━━━━━━━━━━━━━━━━━━ 17s 423ms/step - loss: nan
Epoch 33/50
40/40 ━━━━━━━━━━━━━━━━━━━━ 19s 485ms/step - loss: nan
Epoch 34/50
40/40 ━━━━━━━━━━━━━━━━━━━━ 19s 456ms/step - loss: nan
Epoch 35/50
40/40 ━━━━━━━━━━━━━━━━━━━━ 14s 353ms/step - loss: nan
Epoch 36/50
40/40 ━━━━━━━━━━━━━━━━━━━━ 25s 603ms/step - loss: nan
Epoch 37/50
40/40 ━━━━━━━━━━━━━━━━━━━━ 15s 368ms/step - loss: nan
Epoch 38/50
40/40 ━━━━━━━━━━━━━━━━━━━━ 22s 536ms/step - loss: nan
Epoch 39/50
40/40 ━━━━━━━━━━━━━━━━━━━━ 15s 382ms/step - loss: nan
Epoch 40/50
40/40 ━━━━━━━━━━━━━━━━━━━━ 21s 509ms/step - loss: nan
Epoch 41/50
40/40 ━━━━━━━━━━━━━━━━━━━━ 8s 201ms/step - loss: nan
Epoch 42/50
40/40 ━━━━━━━━━━━━━━━━━━━━ 8s 195ms/step - loss: nan
Epoch 43/50
40/40 ━━━━━━━━━━━━━━━━━━━━ 8s 194ms/step - loss: nan
Epoch 44/50
40/40 ━━━━━━━━━━━━━━━━━━━━ 18s 453ms/step - loss: nan
Epoch 45/50
40/40 ━━━━━━━━━━━━━━━━━━━━ 1:00 8s/step - loss: nan
Epoch 46/50
40/40 ━━━━━━━━━━━━━━━━━━━━ 1:00 195ms/step - loss: nan
Epoch 47/50
40/40 ━━━━━━━━━━━━━━━━━━━━ 1:00 332ms/step - loss: nan
Epoch 48/50
40/40 ━━━━━━━━━━━━━━━━━━━━ 1:00 529ms/step - loss: nan
Epoch 49/50
40/40 ━━━━━━━━━━━━━━━━━━━━ 1:00 402ms/step - loss: nan
Epoch 50/50
40/40 ━━━━━━━━━━━━━━━━━━━━ 1:00 10s/step - loss:  nan

Ниже показано пустое изображение при запуске

Код: Выделить всё

y = inputimg[1]
y = y.reshape(-1, 240, 240, 4)
l = model.predict(y)
mask = np.argmax(l, axis=0)
mask *= 127
plt.axis("off")
plt.imshow(mask)

Я пробовал менять оптимизатор, потери и последнюю функцию активации, более того, мое ядро продолжает зависать без видимой причины даже после изменения размера пакета.
Теперь я попытался использовать модель глубокого обучения с сегментацией Python, чтобы посмотреть, решит ли она проблему, но потеря все равно незначительна.>

Подробнее здесь: https://stackoverflow.com/questions/786 ... ment-brain

1719303372

Anonymous

Некоторый контекст:
На входе есть 4 изображения, и я объединил их в 4 канала, а на выходе — 1 изображение.
Входная форма (образец,240,240,4)
Выходная форма : (sample,240,240,1)
Потеря моей модели за 50 эпох — это Нэн, и точность тоже
код:
предварительная обработка и структурирование модели
[code]import tensorflow.keras as kr
import numpy as np
import matplotlib.pylab as plt
#load data
inputimg = np.load('input_dataset.npy')
outputimg = np.load('output_dataset.npy')

inputimg = np.array(inputimg, dtype=np.float32)
outputimg = np.array(outputimg, dtype=np.float32)
outputimg = np.expand_dims(outputimg, axis=3)

x_train = inputimg[:1126].copy()
y_train = outputimg[:1126].copy()
x_train.shape[0] == y_train.shape[0]
x_eval = inputimg[1126:1189].copy()
y_eval = outputimg[1126:1189].copy()
x_test = inputimg[1189:].copy()
y_test = outputimg[1189:].copy()

print("train target:",y_train.shape,"                test target:",y_test.shape)
print(f"train x{x_test.shape}","                 test x:", x_test.shape)

#shuffle data
np.random.seed(0)
np.random.shuffle(x_train)
np.random.shuffle(y_train)
np.random.shuffle(x_eval)
np.random.shuffle(y_eval)

np.random.shuffle(x_test)
np.random.shuffle(y_test)

from tensorflow.keras import layers
inputs = kr.Input((240,240,4))
x = layers.Conv2D(64, 3, strides=2, activation="relu", padding="same")(inputs)
x = layers.Conv2D(64, 3, activation="relu", padding="same")(x)
x = layers.Conv2D(128, 3, strides=2, activation="relu", padding="same")(x)
x = layers.Conv2D(128, 3, activation="relu", padding="same")(x)
x = layers.Conv2D(256, 3, strides=2, padding="same", activation="relu")(x)
x = layers.Conv2D(256, 3, activation="relu", padding="same")(x)

x = layers.Conv2DTranspose(256, 3, activation="relu", padding="same")(x)
x = layers.Conv2DTranspose(256, 3, activation="relu", padding="same", strides=2)(x)
x = layers.Conv2DTranspose(128, 3, activation="relu", padding="same")(x)
x = layers.Conv2DTranspose(128, 3, activation="relu", padding="same", strides=2)(x)
x = layers.Conv2DTranspose(64, 3, activation="relu", padding="same")(x)
x = layers.Conv2DTranspose(64, 3, activation="relu", padding="same", strides=2)(x)

outputs = layers.Conv2D(4, 3, activation="softmax", padding="same")(x)
model = kr.Model(inputs,outputs)
model.summary()

#the BRATS 2021 task 1 brain tumour segmentation dataset
#https://www.kaggle.com/datasets/dschettler8845/brats-2021-task1/code?datasetId=1541666&sortBy=voteCount
[/code]
выходные данные
[code]train target: (1126, 240, 240, 1)                 test target: (62, 240, 240, 1)
train x(62, 240, 240, 4)                  test x: (62, 240, 240, 4)
Model:  "functional_3"
┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━┓
┃ Layer (type)                    ┃ Output Shape           ┃       Param # ┃
┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━┩
│ input_layer_1 (InputLayer)      │ (None, 240, 240, 4)    │             0 │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ conv2d_7 (Conv2D)               │ (None, 120, 120, 64)   │         2,368 │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ conv2d_8 (Conv2D)               │ (None, 120, 120, 64)   │        36,928 │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ conv2d_9 (Conv2D)               │ (None, 60, 60, 128)    │        73,856 │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ conv2d_10 (Conv2D)              │ (None, 60, 60, 128)    │       147,584 │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ conv2d_11 (Conv2D)              │ (None, 30, 30, 256)    │       295,168 │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ conv2d_12 (Conv2D)              │ (None, 30, 30, 256)    │       590,080 │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ conv2d_transpose_6              │ (None, 30, 30, 256)    │       590,080 │
│ (Conv2DTranspose)               │                        │               │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ conv2d_transpose_7              │ (None, 60, 60, 256)    │       590,080 │
│ (Conv2DTranspose)               │                        │               │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ conv2d_transpose_8              │ (None, 60, 60, 128)    │       295,040 │
│ (Conv2DTranspose)               │                        │               │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ conv2d_transpose_9              │ (None, 120, 120, 128)  │       147,584 │
│ (Conv2DTranspose)               │                        │               │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ conv2d_transpose_10             │ (None, 120, 120, 64)   │        73,792 │
│ (Conv2DTranspose)               │                        │               │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ conv2d_transpose_11             │ (None, 240, 240, 64)   │        36,928 │
│ (Conv2DTranspose)               │                        │               │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ conv2d_13 (Conv2D)              │ (None, 240, 240, 4)    │         2,308 │
└─────────────────────────────────┴────────────────────────┴───────────────┘
Total params: 2,881,796 (10.99 MB)
Trainable params: 2,881,796 (10.99 MB)
Non-trainable params: 0 (0.00 B)
[/code]
при компиляции и обучении:
[code]model.compile(optimizer="Adam", loss="sparse_categorical_crossentropy")

callbacks = [
kr.callbacks.ModelCheckpoint("help.keras",
save_best_only=True)
]

history = model.fit(inputimg,outputimg,
epochs=50,
callbacks=callbacks,)
[/code]
результат компиляции и обучения
[code]Epoch 1/50
WARNING: All log messages before absl::InitializeLog() is called are written to STDERR
I0000 00:00:1719246111.166733  116651 service.cc:145] XLA service 0x7f4470003550 initialized for platform CUDA (this does not guarantee that XLA will be used).  Devices:
I0000 00:00:1719246111.169110  116651 service.cc:153]   StreamExecutor device (0): NVIDIA GeForce RTX 4060 Laptop GPU, Compute Capability 8.9
2024-06-24 23:21:51.376397: I tensorflow/compiler/mlir/tensorflow/utils/dump_mlir_util.cc:268] disabling MLIR crash reproducer, set env var `MLIR_CRASH_REPRODUCER_DIRECTORY` to enable.
W0000 00:00:1719246111.595718  116651 assert_op.cc:38] Ignoring Assert operator compile_loss/sparse_categorical_crossentropy/SparseSoftmaxCrossEntropyWithLogits/assert_equal_1/Assert/Assert
2024-06-24 23:21:51.959587: I external/local_xla/xla/stream_executor/cuda/cuda_dnn.cc:465] Loaded cuDNN version 8907
2024-06-24 23:22:05.935224: E external/local_xla/xla/service/slow_operation_alarm.cc:65] Trying algorithm eng0{} for conv (f32[32,128,121,121]{3,2,1,0}, u8[0]{0}) custom-call(f32[32,128,60,60]{3,2,1,0}, f32[128,128,3,3]{3,2,1,0}), window={size=3x3 stride=2x2}, dim_labels=bf01_oi01->bf01, custom_call_target="__cudnn$convBackwardInput", backend_config={"operation_queue_id":"0","wait_on_operation_queues":[],"cudnn_conv_backend_config":{"conv_result_scale":1,"activation_mode":"kNone","side_input_scale":0,"leakyrelu_alpha":0}} is taking a while...
2024-06-24 23:22:06.654590: E external/local_xla/xla/service/slow_operation_alarm.cc:133] The operation took 1.718804966s
Trying algorithm eng0{} for conv (f32[32,128,121,121]{3,2,1,0}, u8[0]{0}) custom-call(f32[32,128,60,60]{3,2,1,0}, f32[128,128,3,3]{3,2,1,0}), window={size=3x3 stride=2x2}, dim_labels=bf01_oi01->bf01, custom_call_target="__cudnn$convBackwardInput", backend_config={"operation_queue_id":"0","wait_on_operation_queues":[],"cudnn_conv_backend_config":{"conv_result_scale":1,"activation_mode":"kNone","side_input_scale":0,"leakyrelu_alpha":0}} is taking a while...
2024-06-24 23:22:12.322431: E external/local_xla/xla/service/slow_operation_alarm.cc:65] Trying algorithm eng0{} for conv (f32[32,64,241,241]{3,2,1,0}, u8[0]{0}) custom-call(f32[32,64,120,120]{3,2,1,0}, f32[64,64,3,3]{3,2,1,0}), window={size=3x3 stride=2x2}, dim_labels=bf01_oi01->bf01, custom_call_target="__cudnn$convBackwardInput", backend_config={"operation_queue_id":"0","wait_on_operation_queues":[],"cudnn_conv_backend_config":{"conv_result_scale":1,"activation_mode":"kNone","side_input_scale":0,"leakyrelu_alpha":0}} is taking a while...
2024-06-24 23:22:12.953718: E external/local_xla/xla/service/slow_operation_alarm.cc:133] The operation took 1.631474573s
Trying algorithm eng0{} for conv (f32[32,64,241,241]{3,2,1,0}, u8[0]{0}) custom-call(f32[32,64,120,120]{3,2,1,0}, f32[64,64,3,3]{3,2,1,0}), window={size=3x3 stride=2x2}, dim_labels=bf01_oi01->bf01, custom_call_target="__cudnn$convBackwardInput", backend_config={"operation_queue_id":"0","wait_on_operation_queues":[],"cudnn_conv_backend_config":{"conv_result_scale":1,"activation_mode":"kNone","side_input_scale":0,"leakyrelu_alpha":0}} is taking a while...
2024-06-24 23:22:33.411114: E external/local_xla/xla/service/slow_operation_alarm.cc:65] Trying algorithm eng0{} for conv (f32[32,256,60,60]{3,2,1,0}, u8[0]{0}) custom-call(f32[32,128,60,60]{3,2,1,0}, f32[256,128,3,3]{3,2,1,0}), window={size=3x3 pad=1_1x1_1}, dim_labels=bf01_oi01->bf01, custom_call_target="__cudnn$convForward", backend_config={"operation_queue_id":"0","wait_on_operation_queues":[],"cudnn_conv_backend_config":{"conv_result_scale":1,"activation_mode":"kNone","side_input_scale":0,"leakyrelu_alpha":0}} is taking a while...
2024-06-24 23:22:34.596435: E external/local_xla/xla/service/slow_operation_alarm.cc:133] The operation took 2.184202328s
Trying algorithm eng0{} for conv (f32[32,256,60,60]{3,2,1,0}, u8[0]{0}) custom-call(f32[32,128,60,60]{3,2,1,0}, f32[256,128,3,3]{3,2,1,0}), window={size=3x3 pad=1_1x1_1}, dim_labels=bf01_oi01->bf01, custom_call_target="__cudnn$convForward", backend_config={"operation_queue_id":"0","wait_on_operation_queues":[],"cudnn_conv_backend_config":{"conv_result_scale":1,"activation_mode":"kNone","side_input_scale":0,"leakyrelu_alpha":0}} is taking a while...
I0000 00:00:1719246187.497218  116651 device_compiler.h:188] Compiled cluster using XLA!  This line is logged at most once for the lifetime of the process.
39/40 ━━━━━━━━━━━━━━━━━━━━ 0s 370ms/step - loss:  nan
W0000 00:00:1719246202.227021  116653 assert_op.cc:38] Ignoring Assert operator compile_loss/sparse_categorical_crossentropy/SparseSoftmaxCrossEntropyWithLogits/assert_equal_1/Assert/Assert
40/40 ━━━━━━━━━━━━━━━━━━━━ 114s 717ms/step - loss: nan
Epoch 2/50
WARNING: All log messages before absl::InitializeLog() is called are written to STDERR
I0000 00:00:1719246215.883065  116653 asm_compiler.cc:369] ptxas warning : Registers are spilled to local memory in function 'input_reduce_select_fusion_7', 188 bytes spill stores, 188 bytes spill loads
ptxas warning : Registers are spilled to local memory in function 'input_compare_reduce_fusion_1', 228 bytes spill stores, 228 bytes spill loads
ptxas warning : Registers are spilled to local memory in function 'input_reduce_select_fusion_9', 200 bytes spill stores, 200 bytes spill loads

/home/red/.pyenv/versions/3.11.4/lib/python3.11/site-packages/keras/src/callbacks/model_checkpoint.py:206: UserWarning: Can save best model only with val_loss available, skipping.
self._save_model(epoch=epoch, batch=None, logs=logs)
40/40 ━━━━━━━━━━━━━━━━━━━━ 14s 342ms/step - loss: nan
Epoch 3/50
40/40 ━━━━━━━━━━━━━━━━━━━━ 14s 359ms/step - loss: nan
Epoch 4/50
40/40 ━━━━━━━━━━━━━━━━━━━━ 15s 368ms/step - loss: nan
Epoch 5/50
40/40 ━━━━━━━━━━━━━━━━━━━━ 15s 361ms/step - loss: nan
Epoch 6/50
40/40 ━━━━━━━━━━━━━━━━━━━━ 15s 361ms/step - loss: nan
Epoch 7/50
40/40 ━━━━━━━━━━━━━━━━━━━━ 14s 358ms/step - loss: nan
Epoch 8/50
40/40 ━━━━━━━━━━━━━━━━━━━━ 15s 362ms/step - loss: nan
Epoch 9/50
40/40 ━━━━━━━━━━━━━━━━━━━━ 15s 378ms/step - loss: nan
Epoch 10/50
40/40 ━━━━━━━━━━━━━━━━━━━━ 16s 390ms/step - loss: nan
Epoch 11/50
40/40 ━━━━━━━━━━━━━━━━━━━━ 16s 397ms/step - loss: nan
Epoch 12/50
40/40 ━━━━━━━━━━━━━━━━━━━━ 15s 361ms/step - loss: nan
Epoch 13/50
40/40 ━━━━━━━━━━━━━━━━━━━━ 15s 365ms/step - loss: nan
Epoch 14/50
40/40 ━━━━━━━━━━━━━━━━━━━━ 14s 359ms/step - loss: nan
Epoch 15/50
40/40 ━━━━━━━━━━━━━━━━━━━━ 14s 348ms/step - loss: nan
Epoch 16/50
40/40 ━━━━━━━━━━━━━━━━━━━━ 14s 355ms/step - loss: nan
Epoch 17/50
40/40 ━━━━━━━━━━━━━━━━━━━━ 14s 355ms/step - loss: nan
Epoch 18/50
40/40 ━━━━━━━━━━━━━━━━━━━━ 15s 372ms/step - loss: nan
Epoch 19/50
40/40 ━━━━━━━━━━━━━━━━━━━━ 15s 362ms/step - loss: nan
Epoch 20/50
40/40 ━━━━━━━━━━━━━━━━━━━━ 14s 344ms/step - loss: nan
Epoch 21/50
40/40 ━━━━━━━━━━━━━━━━━━━━ 14s 339ms/step - loss: nan
Epoch 22/50
40/40 ━━━━━━━━━━━━━━━━━━━━ 14s 343ms/step - loss: nan
Epoch 23/50
40/40 ━━━━━━━━━━━━━━━━━━━━ 14s 341ms/step - loss: nan
Epoch 24/50
40/40 ━━━━━━━━━━━━━━━━━━━━ 16s 391ms/step - loss: nan
Epoch 25/50
40/40 ━━━━━━━━━━━━━━━━━━━━ 15s 376ms/step - loss: nan
Epoch 26/50
40/40 ━━━━━━━━━━━━━━━━━━━━ 16s 392ms/step - loss: nan
Epoch 27/50
40/40 ━━━━━━━━━━━━━━━━━━━━ 14s 360ms/step - loss: nan
Epoch 28/50
40/40 ━━━━━━━━━━━━━━━━━━━━ 15s 378ms/step - loss: nan
Epoch 29/50
40/40 ━━━━━━━━━━━━━━━━━━━━ 13s 332ms/step - loss: nan
Epoch 30/50
40/40 ━━━━━━━━━━━━━━━━━━━━ 14s 355ms/step - loss: nan
Epoch 31/50
40/40 ━━━━━━━━━━━━━━━━━━━━ 15s 381ms/step - loss: nan
Epoch 32/50
40/40 ━━━━━━━━━━━━━━━━━━━━ 17s 423ms/step - loss: nan
Epoch 33/50
40/40 ━━━━━━━━━━━━━━━━━━━━ 19s 485ms/step - loss: nan
Epoch 34/50
40/40 ━━━━━━━━━━━━━━━━━━━━ 19s 456ms/step - loss: nan
Epoch 35/50
40/40 ━━━━━━━━━━━━━━━━━━━━ 14s 353ms/step - loss: nan
Epoch 36/50
40/40 ━━━━━━━━━━━━━━━━━━━━ 25s 603ms/step - loss: nan
Epoch 37/50
40/40 ━━━━━━━━━━━━━━━━━━━━ 15s 368ms/step - loss: nan
Epoch 38/50
40/40 ━━━━━━━━━━━━━━━━━━━━ 22s 536ms/step - loss: nan
Epoch 39/50
40/40 ━━━━━━━━━━━━━━━━━━━━ 15s 382ms/step - loss: nan
Epoch 40/50
40/40 ━━━━━━━━━━━━━━━━━━━━ 21s 509ms/step - loss: nan
Epoch 41/50
40/40 ━━━━━━━━━━━━━━━━━━━━ 8s 201ms/step - loss: nan
Epoch 42/50
40/40 ━━━━━━━━━━━━━━━━━━━━ 8s 195ms/step - loss: nan
Epoch 43/50
40/40 ━━━━━━━━━━━━━━━━━━━━ 8s 194ms/step - loss: nan
Epoch 44/50
40/40 ━━━━━━━━━━━━━━━━━━━━ 18s 453ms/step - loss: nan
Epoch 45/50
40/40 ━━━━━━━━━━━━━━━━━━━━ 1:00 8s/step - loss: nan
Epoch 46/50
40/40 ━━━━━━━━━━━━━━━━━━━━ 1:00 195ms/step - loss: nan
Epoch 47/50
40/40 ━━━━━━━━━━━━━━━━━━━━ 1:00 332ms/step - loss: nan
Epoch 48/50
40/40 ━━━━━━━━━━━━━━━━━━━━ 1:00 529ms/step - loss: nan
Epoch 49/50
40/40 ━━━━━━━━━━━━━━━━━━━━ 1:00 402ms/step - loss: nan
Epoch 50/50
40/40 ━━━━━━━━━━━━━━━━━━━━ 1:00 10s/step - loss:  nan

[/code]
Ниже показано пустое изображение при запуске
[code]y = inputimg[1]
y = y.reshape(-1, 240, 240, 4)
l = model.predict(y)
mask = np.argmax(l, axis=0)
mask *= 127
plt.axis("off")
plt.imshow(mask)
[/code]
Я пробовал менять оптимизатор, потери и последнюю функцию активации, более того, мое ядро продолжает зависать без видимой причины даже после изменения размера пакета.
Теперь я попытался использовать модель глубокого обучения с сегментацией Python, чтобы посмотреть, решит ли она проблему, но потеря все равно незначительна.> 

Подробнее здесь: [url]https://stackoverflow.com/questions/78665443/nan-loss-in-keras-when-working-on-a-semantic-segmentation-model-to-segment-brain[/url]

Ответить Пред. тема След. тема

1 сообщение • Страница 1 из 1

Быстрый ответ

Заголовок:

Имя пользователя:

Изменение регистра текста:

Смайлики

Ещё смайлики…

К этому ответу прикреплено по крайней мере одно вложение.

Если вы не хотите добавлять вложения, оставьте поля пустыми. Можно прикреплять файлы, перетаскивая их в окно сообщения.

Максимально разрешённый размер вложения: 15 МБ.

Имя файла:

Комментарий к файлу:

Имя файла	Комментарий к файлу	Размер	Статус

Похожие темы

Ответы

Просмотры

Последнее сообщение

Очень низкая точность модели EfficeNetB3 и плато обучения при выявлении проблемы с опухолями головного мозга

Последнее сообщение Anonymous « 27 янв 2025, 14:58
Добавлено в форуме Python

Anonymous » 27 янв 2025, 14:58 » в форуме Python

Ссылка НАБОР ДАННЫХ: (
Это код
# system-libraries
import os
import shutil
import time
import pathlib
import itertools

# data-handling
import cv2
import numpy as np
import pandas as pd
import matplotlib.pyplot as plt
import seaborn as sns...

0 Ответы

12 Просмотры

Последнее сообщение Anonymous
27 янв 2025, 14:58
Очень низкая точность модели EfficeNetB3 и плато обучения при выявлении проблемы с опухолями головного мозга

Последнее сообщение Anonymous « 27 янв 2025, 17:29
Добавлено в форуме Python

Anonymous » 27 янв 2025, 17:29 » в форуме Python

Ссылка НАБОР ДАННЫХ: (
Это код
# system-libraries
import os
import shutil
import time
import pathlib
import itertools

# data-handling
import cv2
import numpy as np
import pandas as pd
import matplotlib.pyplot as plt
import seaborn as sns...

0 Ответы

18 Просмотры

Последнее сообщение Anonymous
27 янв 2025, 17:29
Очень низкая точность модели EfficeNetB3 и плато обучения при выявлении проблемы с опухолями головного мозга

Последнее сообщение Anonymous « 30 янв 2025, 19:54
Добавлено в форуме Python

Anonymous » 30 янв 2025, 19:54 » в форуме Python

Ссылка на набор данных: ( Это код
# system-libraries
import os
import shutil
import time
import pathlib
import itertools

# data-handling
import cv2
import numpy as np
import pandas as pd
import matplotlib.pyplot as plt
import seaborn as sns...

0 Ответы

15 Просмотры

Последнее сообщение Anonymous
30 янв 2025, 19:54
Пытался разработать пользовательский интерфейс для классификации изображений МРТ головного мозга.

Последнее сообщение Anonymous « 03 дек 2024, 12:44
Добавлено в форуме Python

Anonymous » 03 дек 2024, 12:44 » в форуме Python

Я использую блокнот Google Colab Python для разработки пользовательского интерфейса для демонстрационного интерфейса классификации опухолей головного мозга путем импорта потока и запуска файла app.py для интерфейса. Я собираюсь позволить...

0 Ответы

16 Просмотры

Последнее сообщение Anonymous
03 дек 2024, 12:44
Как изменить мой код для обработки изображений RGBX (4-канальных) для семантической сегментации?

Последнее сообщение Anonymous « 18 ноя 2024, 23:06
Добавлено в форуме Python

Anonymous » 18 ноя 2024, 23:06 » в форуме Python

Я новичок в этой области и следую руководству U-Net по использованию трехканальных изображений RGB для семантической сегментации 68HR_eyzk00 и это сработало нормально для меня. Однако теперь мне нужно расширить конвейер для поддержки 4-канальных...

0 Ответы

15 Просмотры

Последнее сообщение Anonymous
18 ноя 2024, 23:06

Вернуться в «Python»