下面是我使用 ResNet50 的 CNN 代码，我过去曾使用过它，但是，我最近更改了模型输入的结构，似乎出现了以下错误。我还打印了图像阵列的形状，您可以在下面看到
train_inputs.shape = (4727, 224, 224, 3)len(train_labels) = 4727此外，错误似乎发生在以下代码块中：
history = model.fit(train_inputs_resized, train_labels, epochs=10, batch_size=32, validation_split=0.2, callbacks=[early_stopping, model_checkpoint])
错误信息：
Packages Loaded
main started
(4727, 6, 100, 100)
4727
images collected
(4727, 6, 224, 224)
(4727, 6, 224, 224)
(4727, 6, 224, 224, 3)
(4727, 224, 224, 3)
model loaded
model built
(4727, 224, 224, 3)
4727
Epoch 1/10
Traceback (most recent call last):
File "/oscar/scratch/aagudel1/SenID/Code/CNN_RN503D.py", line 388, in <module>
main()
File "/oscar/scratch/aagudel1/SenID/Code/CNN_RN503D.py", line 352, in main
history = model.fit(train_inputs_resized, train_labels, epochs=10, batch_siz e=32, validation_split=0.2, callbacks=[early_stopping, model_checkpoint])
File "/users/aagudel1/.local/lib/python3.10/site-packages/keras/src/utils/trac eback_utils.py", line 122, in error_handler
raise e.with_traceback(filtered_tb) from None
File "/users/aagudel1/.local/lib/python3.10/site-packages/tensorflow/python/ea ger/execute.py", line 53, in quick_execute
tensors = pywrap_tfe.TFE_Py_Execute(ctx._handle, device_name, op_name,
tensorflow.python.framework.errors_impl.InvalidArgumentError: Graph execution er ror:
Detected at node LogicalAnd defined at (most recent call last):
File "/oscar/scratch/aagudel1/SenID/Code/CNN_RN503D.py", line 388, in <module>
File "/oscar/scratch/aagudel1/SenID/Code/CNN_RN503D.py", line 352, in main
File "/users/aagudel1/.local/lib/python3.10/site-packages/keras/src/utils/traceback_utils.py", line 117, in error_handler
File "/users/aagudel1/.local/lib/python3.10/site-packages/keras/src/backend/tensorflow/trainer.py", line 325, in fit
File "/users/aagudel1/.local/lib/python3.10/site-packages/keras/src/backend/tensorflow/trainer.py", line 118, in one_step_on_iterator
File "/users/aagudel1/.local/lib/python3.10/site-packages/keras/src/backend/tensorflow/trainer.py", line 106, in one_step_on_data
File "/users/aagudel1/.local/lib/python3.10/site-packages/keras/src/backend/tensorflow/trainer.py", line 77, in train_step
File "/users/aagudel1/.local/lib/python3.10/site-packages/keras/src/trainers/trainer.py", line 376, in compute_metrics
File "/users/aagudel1/.local/lib/python3.10/site-packages/keras/src/traineompile_utils.py", line 330, in update_state
File "/users/aagudel1/.local/lib/python3.10/site-packages/keras/src/trompile_utils.py", line 17, in update_state
File "/users/aagudel1/.local/lib/python3.10/site-packages/keras/src/bansorflow/numpy.py", line 1194, in logical_and
Incompatible shapes: [1,64] vs. [1,32]
[[{{node LogicalAnd}}]] [Op:__inference_one_step_on_iterator_12
主要功能完整代码：
def main():
"""
This script performs the following tasks:
1. Collects and processes images of cycling and senescent cells.
2. Splits the data into training and testing sets.
3. Loads a pre-trained ResNet50 model and adds custom layers.
4. Trains the model using the processed images.
5. Evaluates the model's performance on the test set.
6. Generates a confusion matrix and classification report.
7. Visualizes the results, including correctly and incorrectly classified examples.
"""
print("main started")
script_dir = os.path.dirname(os.path.abspath(__file__))
output_dir = os.path.join(script_dir, "../Data/")
out1_dir = os.path.join(output_dir, "/users/aagudel1/scratch/SenID/Images/SegmentedImages/Prolif/")
out2_dir = os.path.join(output_dir, "/users/aagudel1/scratch/SenID/Images/SegmentedImages/Senescent/")
train_inputs, test_inputs, train_labels, test_labels = collect_images(out1_dir, out2_dir, train_ratio=0.8, random_state=42)
print(train_inputs.shape)
print(len(train_labels))
print("images collected")
# Assuming images are already 3D
train_inputs = ImageCropper.crop_images(train_inputs)
print(train_inputs.shape)
test_inputs = ImageCropper.crop_images(test_inputs)
train_inputs = preprocess_input(train_inputs)
test_inputs = preprocess_input(test_inputs)
print(train_inputs.shape)
train_inputs_resized = np.repeat(train_inputs[..., np.newaxis], 3, axis=-1)
print(train_inputs_resized.shape)
test_inputs_resized = np.repeat(test_inputs[..., np.newaxis], 3, axis=-1)
# Combine slices using Maximum Intensity Projection (MIP)
train_inputs_resized = np.max(train_inputs_resized, axis=1)
print(train_inputs_resized.shape)
test_inputs_resized = np.max(test_inputs_resized, axis=1)
# Load ResNet50 as a base model and add custom layers
base_model = ResNet50(weights='imagenet', include_top=False)
print("model loaded")
# Unfreeze the last few layers
for layer in base_model.layers[:-5]:
layer.trainable = False
for layer in base_model.layers[-5:]:
layer.trainable = True
# Add custom layers
x = base_model.output
x = GlobalAveragePooling2D()(x)
x = Dense(256, activation='relu')(x)
x = Dropout(0.5)(x)
predictions = Dense(2, activation='softmax', kernel_regularizer=regularizers.l2(0.01))(x)
# Build the final model
model = Model(inputs=base_model.input, outputs=predictions)
print("model built")
# Compile the model
model.compile(optimizer='adam', loss='sparse_categorical_crossentropy', metrics=['accuracy', Precision(), Recall()])
# Early stopping and model checkpoints
early_stopping = EarlyStopping(monitor='val_loss', patience=5)
model_checkpoint = ModelCheckpoint('best_model.keras', monitor='val_loss', save_best_only=True)
# Train the model
print(train_inputs_resized.shape)
print(len(train_labels))
history = model.fit(train_inputs_resized, train_labels, epochs=10, batch_size=32, validation_split=0.2, callbacks=[early_stopping, model_checkpoint])
plot_history(history, "model_loss.png")
print("model trained")
#tf.print("Value of output tensor:", output)
# Evaluate the model
test_loss, test_acc = model.evaluate(test_inputs_resized, test_labels)
print(f"Test accuracy: {test_acc}")
print(f"Test loss: {test_loss}")
y_pred = model.predict(test_inputs_resized)
y_pred_binary = np.argmax(y_pred, axis=1)
class_labels = ["Cycling", "Senescene"]
create_confusion_matrix(test_labels, y_pred_binary, class_labels, "confusion_matrix.png")
print(classification_report(test_labels, y_pred_binary))
accuracy = accuracy_score(test_labels, y_pred_binary)
print("Test Accuracy", accuracy.round(4) * 100)
precision = precision_score(test_labels, y_pred_binary)
print("Test Precision", precision.round(4) * 100)
recall = recall_score(test_labels, y_pred_binary)
print("Test Recall", recall.round(4) * 100)
f1 = f1_score(test_labels, y_pred_binary)
print("Test F1", f1.round(4) * 100)
test_inputs = (test_inputs_resized - test_inputs_resized.min()) / (test_inputs_resized.max() - test_inputs_resized.min())
visualize_results(
image_inputs=test_inputs_resized,
probabilities=y_pred,
image_labels=test_labels,
first_label="Cycling",
second_label="Senescene")
关于如何解决此问题或排除故障有什么想法吗？错误消息显示形状 [1,64] 与 [1,32] 不兼容，但我不知道它可能指的是哪些形状
如果你只有 2 门课，最好使用
loss = 'binary_crossentropy'Dense(1, active='sigmoid', ...)