Concrete Crack Detector

Run in Google Colab

Objective: Use transfer learning to build an image classifier that detects concrete with and without cracks.

Pre-trained models such as VGG-16, Inception, and ResNet can be used for image classification. These models are trained on large datasets like ImageNet, which contains over a million images and can classify images into 1000 object categories. By leveraging these pre-trained models, we can fine-tune them for specific tasks with smaller datasets, improving accuracy and reducing the need for extensive training data.

Import libraries

import os
import tensorflow as tf
from keras import utils, applications, Input, layers, Model
from sklearn.metrics import classification_report, ConfusionMatrixDisplay
import matplotlib.pyplot as plt
import seaborn as sns
sns.set_style("whitegrid")

2025-03-07 02:42:29.359798: E external/local_xla/xla/stream_executor/cuda/cuda_fft.cc:477] Unable to register cuFFT factory: Attempting to register factory for plugin cuFFT when one has already been registered
WARNING: All log messages before absl::InitializeLog() is called are written to STDERR
E0000 00:00:1741336949.372462  462140 cuda_dnn.cc:8310] Unable to register cuDNN factory: Attempting to register factory for plugin cuDNN when one has already been registered
E0000 00:00:1741336949.376536  462140 cuda_blas.cc:1418] Unable to register cuBLAS factory: Attempting to register factory for plugin cuBLAS when one has already been registered
2025-03-07 02:42:29.389194: I tensorflow/core/platform/cpu_feature_guard.cc:210] This TensorFlow binary is optimized to use available CPU instructions in performance-critical operations.
To enable the following instructions: SSE4.1 SSE4.2 AVX AVX2 FMA, in other operations, rebuild TensorFlow with the appropriate compiler flags.

Download the dataset

%%bash

wget -nc --progress=bar:force:noscroll https://cf-courses-data.s3.us.cloud-object-storage.appdomain.cloud/IBMDeveloperSkillsNetwork-DL0321EN-SkillsNetwork/concrete_data_week3.zip -P /tmp
unzip -qn /tmp/concrete_data_week3.zip -d /tmp

File '/tmp/concrete_data_week3.zip' already there; not retrieving.

Load the dataset

image_size = (64, 64) # Reduce image size to speed up training and reduce memory consumption
batch_size = 100

ds_train = utils.image_dataset_from_directory(directory='/tmp/concrete_data_week3/train',
                                              image_size=image_size,
                                              batch_size=batch_size)

ds_validation, ds_test = utils.image_dataset_from_directory(directory='/tmp/concrete_data_week3/valid',
                                                            image_size=image_size,
                                                            batch_size=batch_size,
                                                            seed=0,
                                                            subset="both",
                                                            validation_split=0.5)

classes = ds_train.class_names

Found 10001 files belonging to 2 classes.

I0000 00:00:1741336952.146354  462140 gpu_device.cc:2022] Created device /job:localhost/replica:0/task:0/device:GPU:0 with 1758 MB memory:  -> device: 0, name: NVIDIA GeForce GTX 1650, pci bus id: 0000:01:00.0, compute capability: 7.5

Found 5001 files belonging to 2 classes.
Using 2501 files for training.
Using 2500 files for validation.

Visualize the dataset

The negative class represents concrete images with no cracks and the positive class represents concrete images with cracks.

images, labels = next(iter(ds_train))

fig, axs = plt.subplots(4, 4, figsize=(8, 8))
fig.suptitle('Random samples')

for i, ax in enumerate(axs.flatten()):
    ax.imshow(images[i] / 255)
    ax.set_title(classes[labels[i].numpy()])
    ax.axis("off")

plt.tight_layout()
plt.show()

Visualize the class distribution

_, _, negative_files = next(os.walk("/tmp/concrete_data_week3/train/negative"))
_, _, positive_files = next(os.walk("/tmp/concrete_data_week3/train/positive"))
sizes = [len(negative_files), len(positive_files)]

fig, ax = plt.subplots()
ax.pie(sizes, textprops={'color': "w", 'fontsize': '12'}, autopct=lambda pct: "{:.2f}%\n({:d})".format(pct, round(pct/100 * sum(sizes))))
ax.legend(["0 (negative)", "1 (positive)"])
ax.set_title("Class")
plt.show()

Build a CNN using the ResNet50 architecture as a base

input_shape = (64, 64, 3)
ResNet50 = applications.ResNet50(include_top=False, input_shape=input_shape, pooling="avg")
ResNet50.trainable = False

inputs = Input(shape=input_shape)
x = applications.resnet.preprocess_input(inputs) # Preprocess our images the same way the images used to train ResNet50 model were processed
x = ResNet50(x, training=False)
outputs = layers.Dense(1, activation="sigmoid")(x)

model_resnet = Model(inputs, outputs)
model_resnet.summary()

Model: "functional"

┏━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━┓
┃ Layer (type)        ┃ Output Shape      ┃    Param # ┃ Connected to      ┃
┡━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━┩
│ input_layer_1       │ (None, 64, 64, 3) │          0 │ -                 │
│ (InputLayer)        │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ get_item (GetItem)  │ (None, 64, 64)    │          0 │ input_layer_1[0]… │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ get_item_1          │ (None, 64, 64)    │          0 │ input_layer_1[0]… │
│ (GetItem)           │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ get_item_2          │ (None, 64, 64)    │          0 │ input_layer_1[0]… │
│ (GetItem)           │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ stack (Stack)       │ (None, 64, 64, 3) │          0 │ get_item[0][0],   │
│                     │                   │            │ get_item_1[0][0], │
│                     │                   │            │ get_item_2[0][0]  │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ add (Add)           │ (None, 64, 64, 3) │          0 │ stack[0][0]       │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ resnet50            │ (None, 2048)      │ 23,587,712 │ add[0][0]         │
│ (Functional)        │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ dense (Dense)       │ (None, 1)         │      2,049 │ resnet50[0][0]    │
└─────────────────────┴───────────────────┴────────────┴───────────────────┘

 Total params: 23,589,761 (89.99 MB)

 Trainable params: 2,049 (8.00 KB)

 Non-trainable params: 23,587,712 (89.98 MB)

Compile and train the ResNet50 model

model_resnet.compile(optimizer="adam", loss='binary_crossentropy', metrics=['binary_accuracy'])

epochs = 5
history_resnet = model_resnet.fit(ds_train, epochs=epochs, validation_data=ds_validation)

Epoch 1/5

I0000 00:00:1741336960.082890  462220 service.cc:148] XLA service 0x7f6cb8001e90 initialized for platform CUDA (this does not guarantee that XLA will be used). Devices:
I0000 00:00:1741336960.082915  462220 service.cc:156]   StreamExecutor device (0): NVIDIA GeForce GTX 1650, Compute Capability 7.5
2025-03-07 02:42:40.218994: I tensorflow/compiler/mlir/tensorflow/utils/dump_mlir_util.cc:268] disabling MLIR crash reproducer, set env var `MLIR_CRASH_REPRODUCER_DIRECTORY` to enable.
I0000 00:00:1741336961.058020  462220 cuda_dnn.cc:529] Loaded cuDNN version 90701
2025-03-07 02:42:41.463080: I external/local_xla/xla/service/gpu/autotuning/conv_algorithm_picker.cc:557] Omitted potentially buggy algorithm eng14{k25=2} for conv (f32[100,64,16,16]{3,2,1,0}, u8[0]{0}) custom-call(f32[100,64,16,16]{3,2,1,0}, f32[64,64,3,3]{3,2,1,0}, f32[64]{0}), window={size=3x3 pad=1_1x1_1}, dim_labels=bf01_oi01->bf01, custom_call_target="__cudnn$convBiasActivationForward", backend_config={"cudnn_conv_backend_config":{"activation_mode":"kNone","conv_result_scale":1,"leakyrelu_alpha":0,"side_input_scale":0},"force_earliest_schedule":false,"operation_queue_id":"0","wait_on_operation_queues":[]}
2025-03-07 02:42:41.739413: I external/local_xla/xla/service/gpu/autotuning/conv_algorithm_picker.cc:557] Omitted potentially buggy algorithm eng14{k25=2} for conv (f32[100,128,8,8]{3,2,1,0}, u8[0]{0}) custom-call(f32[100,128,8,8]{3,2,1,0}, f32[128,128,3,3]{3,2,1,0}, f32[128]{0}), window={size=3x3 pad=1_1x1_1}, dim_labels=bf01_oi01->bf01, custom_call_target="__cudnn$convBiasActivationForward", backend_config={"cudnn_conv_backend_config":{"activation_mode":"kNone","conv_result_scale":1,"leakyrelu_alpha":0,"side_input_scale":0},"force_earliest_schedule":false,"operation_queue_id":"0","wait_on_operation_queues":[]}
2025-03-07 02:42:41.997736: I external/local_xla/xla/service/gpu/autotuning/conv_algorithm_picker.cc:557] Omitted potentially buggy algorithm eng14{k25=2} for conv (f32[100,256,4,4]{3,2,1,0}, u8[0]{0}) custom-call(f32[100,256,4,4]{3,2,1,0}, f32[256,256,3,3]{3,2,1,0}, f32[256]{0}), window={size=3x3 pad=1_1x1_1}, dim_labels=bf01_oi01->bf01, custom_call_target="__cudnn$convBiasActivationForward", backend_config={"cudnn_conv_backend_config":{"activation_mode":"kNone","conv_result_scale":1,"leakyrelu_alpha":0,"side_input_scale":0},"force_earliest_schedule":false,"operation_queue_id":"0","wait_on_operation_queues":[]}
2025-03-07 02:42:42.393243: W external/local_xla/xla/tsl/framework/bfc_allocator.cc:306] Allocator (GPU_0_bfc) ran out of memory trying to allocate 1.49GiB with freed_by_count=0. The caller indicates that this is not a failure, but this may mean that there could be performance gains if more memory were available.

  3/101 ━━━━━━━━━━━━━━━━━━━━ 4s 44ms/step - binary_accuracy: 0.7694 - loss: 0.5314 

I0000 00:00:1741336963.422988  462220 device_compiler.h:188] Compiled cluster using XLA!  This line is logged at most once for the lifetime of the process.

 99/101 ━━━━━━━━━━━━━━━━━━━━ 0s 39ms/step - binary_accuracy: 0.9420 - loss: 0.1842

2025-03-07 02:42:48.362201: I external/local_xla/xla/service/gpu/autotuning/conv_algorithm_picker.cc:557] Omitted potentially buggy algorithm eng14{k25=2} for conv (f32[1,64,16,16]{3,2,1,0}, u8[0]{0}) custom-call(f32[1,64,16,16]{3,2,1,0}, f32[64,64,3,3]{3,2,1,0}, f32[64]{0}), window={size=3x3 pad=1_1x1_1}, dim_labels=bf01_oi01->bf01, custom_call_target="__cudnn$convBiasActivationForward", backend_config={"cudnn_conv_backend_config":{"activation_mode":"kNone","conv_result_scale":1,"leakyrelu_alpha":0,"side_input_scale":0},"force_earliest_schedule":false,"operation_queue_id":"0","wait_on_operation_queues":[]}
2025-03-07 02:42:48.493109: I external/local_xla/xla/service/gpu/autotuning/conv_algorithm_picker.cc:557] Omitted potentially buggy algorithm eng14{k25=2} for conv (f32[1,128,8,8]{3,2,1,0}, u8[0]{0}) custom-call(f32[1,128,8,8]{3,2,1,0}, f32[128,128,3,3]{3,2,1,0}, f32[128]{0}), window={size=3x3 pad=1_1x1_1}, dim_labels=bf01_oi01->bf01, custom_call_target="__cudnn$convBiasActivationForward", backend_config={"cudnn_conv_backend_config":{"activation_mode":"kNone","conv_result_scale":1,"leakyrelu_alpha":0,"side_input_scale":0},"force_earliest_schedule":false,"operation_queue_id":"0","wait_on_operation_queues":[]}
2025-03-07 02:42:48.650442: I external/local_xla/xla/service/gpu/autotuning/conv_algorithm_picker.cc:557] Omitted potentially buggy algorithm eng14{k25=2} for conv (f32[1,256,4,4]{3,2,1,0}, u8[0]{0}) custom-call(f32[1,256,4,4]{3,2,1,0}, f32[256,256,3,3]{3,2,1,0}, f32[256]{0}), window={size=3x3 pad=1_1x1_1}, dim_labels=bf01_oi01->bf01, custom_call_target="__cudnn$convBiasActivationForward", backend_config={"cudnn_conv_backend_config":{"activation_mode":"kNone","conv_result_scale":1,"leakyrelu_alpha":0,"side_input_scale":0},"force_earliest_schedule":false,"operation_queue_id":"0","wait_on_operation_queues":[]}

101/101 ━━━━━━━━━━━━━━━━━━━━ 19s 112ms/step - binary_accuracy: 0.9430 - loss: 0.1814 - val_binary_accuracy: 0.9940 - val_loss: 0.0308
Epoch 2/5
101/101 ━━━━━━━━━━━━━━━━━━━━ 5s 49ms/step - binary_accuracy: 0.9962 - loss: 0.0193 - val_binary_accuracy: 0.9964 - val_loss: 0.0203
Epoch 3/5
101/101 ━━━━━━━━━━━━━━━━━━━━ 5s 50ms/step - binary_accuracy: 0.9979 - loss: 0.0115 - val_binary_accuracy: 0.9968 - val_loss: 0.0164
Epoch 4/5
101/101 ━━━━━━━━━━━━━━━━━━━━ 5s 49ms/step - binary_accuracy: 0.9985 - loss: 0.0084 - val_binary_accuracy: 0.9968 - val_loss: 0.0143
Epoch 5/5
101/101 ━━━━━━━━━━━━━━━━━━━━ 5s 49ms/step - binary_accuracy: 0.9987 - loss: 0.0066 - val_binary_accuracy: 0.9968 - val_loss: 0.0131

fig, (ax1, ax2) = plt.subplots(1, 2, figsize=(15, 5))

ax1.plot(history_resnet.history['binary_accuracy'])
ax1.plot(history_resnet.history['val_binary_accuracy'])
ax1.set_xticks(range(epochs))
ax1.set_xlabel("Epochs")
ax1.set_ylabel("Accuracy")
ax1.legend(["Training", "Validation"])

ax2.plot(history_resnet.history['loss'])
ax2.plot(history_resnet.history['val_loss'])
ax2.set_xticks(range(epochs))
ax2.set_xlabel("Epochs")
ax2.set_ylabel("Loss")
ax2.legend(["Training", "Validation"])

plt.show()

Evaluate the ResNet50 model

images, labels = next(iter(ds_test))

fig, axs = plt.subplots(4, 4, figsize=(8, 8))
fig.suptitle('Random samples')

for i, ax in enumerate(axs.flatten()):
    prediction_proba = model_resnet.predict(tf.expand_dims(images[i], axis=0), verbose=0)
    ax.imshow(images[i] / 255)
    ax.set_title("Prediction: " + classes[int(prediction_proba.squeeze().round())])
    ax.axis("off")

plt.tight_layout()
plt.show()

_, y_test = zip(*ds_test.unbatch())
y_test = [y.numpy() for y in y_test]

prediction_proba = model_resnet.predict(ds_test, verbose=0)
y_pred = prediction_proba.squeeze().round()

print(classification_report(y_test, y_pred, digits=4))

ConfusionMatrixDisplay.from_predictions(y_test, y_pred, display_labels=["0 (negative)", "1 (positive)"])
plt.grid(False)
plt.show()

2025-03-07 02:43:19.602946: I tensorflow/core/framework/local_rendezvous.cc:405] Local rendezvous is aborting with status: OUT_OF_RANGE: End of sequence

              precision    recall  f1-score   support

           0     0.9960    0.9984    0.9972      1249
           1     0.9984    0.9960    0.9972      1251

    accuracy                         0.9972      2500
   macro avg     0.9972    0.9972    0.9972      2500
weighted avg     0.9972    0.9972    0.9972      2500

Build a CNN using the VGG16 architecture as a base

input_shape = (64, 64, 3)
VGG16 = applications.VGG16(include_top=False, input_shape=input_shape, pooling="avg")
VGG16.trainable = False

inputs = Input(shape=input_shape)
x = applications.vgg16.preprocess_input(inputs) # Preprocess our images the same way the images used to train VGG16 model were processed
x = VGG16(x, training=False)
outputs = layers.Dense(1, activation="sigmoid")(x)

model_vgg = Model(inputs, outputs)
model_vgg.summary()

Model: "functional_1"

┏━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━┓
┃ Layer (type)        ┃ Output Shape      ┃    Param # ┃ Connected to      ┃
┡━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━┩
│ input_layer_3       │ (None, 64, 64, 3) │          0 │ -                 │
│ (InputLayer)        │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ get_item_3          │ (None, 64, 64)    │          0 │ input_layer_3[0]… │
│ (GetItem)           │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ get_item_4          │ (None, 64, 64)    │          0 │ input_layer_3[0]… │
│ (GetItem)           │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ get_item_5          │ (None, 64, 64)    │          0 │ input_layer_3[0]… │
│ (GetItem)           │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ stack_1 (Stack)     │ (None, 64, 64, 3) │          0 │ get_item_3[0][0], │
│                     │                   │            │ get_item_4[0][0], │
│                     │                   │            │ get_item_5[0][0]  │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ add_1 (Add)         │ (None, 64, 64, 3) │          0 │ stack_1[0][0]     │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ vgg16 (Functional)  │ (None, 512)       │ 14,714,688 │ add_1[0][0]       │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ dense_1 (Dense)     │ (None, 1)         │        513 │ vgg16[0][0]       │
└─────────────────────┴───────────────────┴────────────┴───────────────────┘

 Total params: 14,715,201 (56.13 MB)

 Trainable params: 513 (2.00 KB)

 Non-trainable params: 14,714,688 (56.13 MB)

Compile and train the VGG16 model

model_vgg.compile(optimizer="adam", loss='binary_crossentropy', metrics=['binary_accuracy'])

epochs = 5
history_vgg = model_vgg.fit(ds_train, epochs=epochs, validation_data=ds_validation)

Epoch 1/5

2025-03-07 02:43:24.550424: I external/local_xla/xla/service/gpu/autotuning/conv_algorithm_picker.cc:557] Omitted potentially buggy algorithm eng14{k25=2} for conv (f32[100,64,64,64]{3,2,1,0}, u8[0]{0}) custom-call(f32[100,3,64,64]{3,2,1,0}, f32[64,3,3,3]{3,2,1,0}, f32[64]{0}), window={size=3x3 pad=1_1x1_1}, dim_labels=bf01_oi01->bf01, custom_call_target="__cudnn$convBiasActivationForward", backend_config={"cudnn_conv_backend_config":{"activation_mode":"kRelu","conv_result_scale":1,"leakyrelu_alpha":0,"side_input_scale":0},"force_earliest_schedule":false,"operation_queue_id":"0","wait_on_operation_queues":[]}
2025-03-07 02:43:24.739802: I external/local_xla/xla/service/gpu/autotuning/conv_algorithm_picker.cc:557] Omitted potentially buggy algorithm eng14{k25=2} for conv (f32[100,64,64,64]{3,2,1,0}, u8[0]{0}) custom-call(f32[100,64,64,64]{3,2,1,0}, f32[64,64,3,3]{3,2,1,0}, f32[64]{0}), window={size=3x3 pad=1_1x1_1}, dim_labels=bf01_oi01->bf01, custom_call_target="__cudnn$convBiasActivationForward", backend_config={"cudnn_conv_backend_config":{"activation_mode":"kRelu","conv_result_scale":1,"leakyrelu_alpha":0,"side_input_scale":0},"force_earliest_schedule":false,"operation_queue_id":"0","wait_on_operation_queues":[]}
2025-03-07 02:43:25.135193: W external/local_xla/xla/tsl/framework/bfc_allocator.cc:306] Allocator (GPU_0_bfc) ran out of memory trying to allocate 1.32GiB with freed_by_count=0. The caller indicates that this is not a failure, but this may mean that there could be performance gains if more memory were available.
2025-03-07 02:43:25.659937: I external/local_xla/xla/service/gpu/autotuning/conv_algorithm_picker.cc:557] Omitted potentially buggy algorithm eng14{k25=2} for conv (f32[100,128,32,32]{3,2,1,0}, u8[0]{0}) custom-call(f32[100,64,32,32]{3,2,1,0}, f32[128,64,3,3]{3,2,1,0}, f32[128]{0}), window={size=3x3 pad=1_1x1_1}, dim_labels=bf01_oi01->bf01, custom_call_target="__cudnn$convBiasActivationForward", backend_config={"cudnn_conv_backend_config":{"activation_mode":"kRelu","conv_result_scale":1,"leakyrelu_alpha":0,"side_input_scale":0},"force_earliest_schedule":false,"operation_queue_id":"0","wait_on_operation_queues":[]}
2025-03-07 02:43:26.122903: I external/local_xla/xla/service/gpu/autotuning/conv_algorithm_picker.cc:557] Omitted potentially buggy algorithm eng14{k25=2} for conv (f32[100,128,32,32]{3,2,1,0}, u8[0]{0}) custom-call(f32[100,128,32,32]{3,2,1,0}, f32[128,128,3,3]{3,2,1,0}, f32[128]{0}), window={size=3x3 pad=1_1x1_1}, dim_labels=bf01_oi01->bf01, custom_call_target="__cudnn$convBiasActivationForward", backend_config={"cudnn_conv_backend_config":{"activation_mode":"kRelu","conv_result_scale":1,"leakyrelu_alpha":0,"side_input_scale":0},"force_earliest_schedule":false,"operation_queue_id":"0","wait_on_operation_queues":[]}
2025-03-07 02:43:26.807555: I external/local_xla/xla/service/gpu/autotuning/conv_algorithm_picker.cc:557] Omitted potentially buggy algorithm eng14{k25=2} for conv (f32[100,256,16,16]{3,2,1,0}, u8[0]{0}) custom-call(f32[100,128,16,16]{3,2,1,0}, f32[256,128,3,3]{3,2,1,0}, f32[256]{0}), window={size=3x3 pad=1_1x1_1}, dim_labels=bf01_oi01->bf01, custom_call_target="__cudnn$convBiasActivationForward", backend_config={"cudnn_conv_backend_config":{"activation_mode":"kRelu","conv_result_scale":1,"leakyrelu_alpha":0,"side_input_scale":0},"force_earliest_schedule":false,"operation_queue_id":"0","wait_on_operation_queues":[]}
2025-03-07 02:43:27.182521: I external/local_xla/xla/service/gpu/autotuning/conv_algorithm_picker.cc:557] Omitted potentially buggy algorithm eng14{k25=2} for conv (f32[100,256,16,16]{3,2,1,0}, u8[0]{0}) custom-call(f32[100,256,16,16]{3,2,1,0}, f32[256,256,3,3]{3,2,1,0}, f32[256]{0}), window={size=3x3 pad=1_1x1_1}, dim_labels=bf01_oi01->bf01, custom_call_target="__cudnn$convBiasActivationForward", backend_config={"cudnn_conv_backend_config":{"activation_mode":"kRelu","conv_result_scale":1,"leakyrelu_alpha":0,"side_input_scale":0},"force_earliest_schedule":false,"operation_queue_id":"0","wait_on_operation_queues":[]}
2025-03-07 02:43:27.764870: I external/local_xla/xla/service/gpu/autotuning/conv_algorithm_picker.cc:557] Omitted potentially buggy algorithm eng14{k25=2} for conv (f32[100,512,8,8]{3,2,1,0}, u8[0]{0}) custom-call(f32[100,256,8,8]{3,2,1,0}, f32[512,256,3,3]{3,2,1,0}, f32[512]{0}), window={size=3x3 pad=1_1x1_1}, dim_labels=bf01_oi01->bf01, custom_call_target="__cudnn$convBiasActivationForward", backend_config={"cudnn_conv_backend_config":{"activation_mode":"kRelu","conv_result_scale":1,"leakyrelu_alpha":0,"side_input_scale":0},"force_earliest_schedule":false,"operation_queue_id":"0","wait_on_operation_queues":[]}
2025-03-07 02:43:28.213943: I external/local_xla/xla/service/gpu/autotuning/conv_algorithm_picker.cc:557] Omitted potentially buggy algorithm eng14{k25=2} for conv (f32[100,512,8,8]{3,2,1,0}, u8[0]{0}) custom-call(f32[100,512,8,8]{3,2,1,0}, f32[512,512,3,3]{3,2,1,0}, f32[512]{0}), window={size=3x3 pad=1_1x1_1}, dim_labels=bf01_oi01->bf01, custom_call_target="__cudnn$convBiasActivationForward", backend_config={"cudnn_conv_backend_config":{"activation_mode":"kRelu","conv_result_scale":1,"leakyrelu_alpha":0,"side_input_scale":0},"force_earliest_schedule":false,"operation_queue_id":"0","wait_on_operation_queues":[]}
2025-03-07 02:43:28.741988: W external/local_xla/xla/tsl/framework/bfc_allocator.cc:306] Allocator (GPU_0_bfc) ran out of memory trying to allocate 1.49GiB with freed_by_count=0. The caller indicates that this is not a failure, but this may mean that there could be performance gains if more memory were available.
2025-03-07 02:43:28.744689: I external/local_xla/xla/service/gpu/autotuning/conv_algorithm_picker.cc:557] Omitted potentially buggy algorithm eng14{k25=2} for conv (f32[100,512,4,4]{3,2,1,0}, u8[0]{0}) custom-call(f32[100,512,4,4]{3,2,1,0}, f32[512,512,3,3]{3,2,1,0}, f32[512]{0}), window={size=3x3 pad=1_1x1_1}, dim_labels=bf01_oi01->bf01, custom_call_target="__cudnn$convBiasActivationForward", backend_config={"cudnn_conv_backend_config":{"activation_mode":"kRelu","conv_result_scale":1,"leakyrelu_alpha":0,"side_input_scale":0},"force_earliest_schedule":false,"operation_queue_id":"0","wait_on_operation_queues":[]}
2025-03-07 02:43:29.006076: W external/local_xla/xla/tsl/framework/bfc_allocator.cc:306] Allocator (GPU_0_bfc) ran out of memory trying to allocate 1.49GiB with freed_by_count=0. The caller indicates that this is not a failure, but this may mean that there could be performance gains if more memory were available.

100/101 ━━━━━━━━━━━━━━━━━━━━ 0s 89ms/step - binary_accuracy: 0.8105 - loss: 0.5264

2025-03-07 02:43:38.527962: I external/local_xla/xla/service/gpu/autotuning/conv_algorithm_picker.cc:557] Omitted potentially buggy algorithm eng14{k25=2} for conv (f32[1,64,64,64]{3,2,1,0}, u8[0]{0}) custom-call(f32[1,64,64,64]{3,2,1,0}, f32[64,64,3,3]{3,2,1,0}, f32[64]{0}), window={size=3x3 pad=1_1x1_1}, dim_labels=bf01_oi01->bf01, custom_call_target="__cudnn$convBiasActivationForward", backend_config={"cudnn_conv_backend_config":{"activation_mode":"kRelu","conv_result_scale":1,"leakyrelu_alpha":0,"side_input_scale":0},"force_earliest_schedule":false,"operation_queue_id":"0","wait_on_operation_queues":[]}
2025-03-07 02:43:38.609825: I external/local_xla/xla/service/gpu/autotuning/conv_algorithm_picker.cc:557] Omitted potentially buggy algorithm eng14{k25=2} for conv (f32[1,128,32,32]{3,2,1,0}, u8[0]{0}) custom-call(f32[1,64,32,32]{3,2,1,0}, f32[128,64,3,3]{3,2,1,0}, f32[128]{0}), window={size=3x3 pad=1_1x1_1}, dim_labels=bf01_oi01->bf01, custom_call_target="__cudnn$convBiasActivationForward", backend_config={"cudnn_conv_backend_config":{"activation_mode":"kRelu","conv_result_scale":1,"leakyrelu_alpha":0,"side_input_scale":0},"force_earliest_schedule":false,"operation_queue_id":"0","wait_on_operation_queues":[]}
2025-03-07 02:43:38.680205: I external/local_xla/xla/service/gpu/autotuning/conv_algorithm_picker.cc:557] Omitted potentially buggy algorithm eng14{k25=2} for conv (f32[1,128,32,32]{3,2,1,0}, u8[0]{0}) custom-call(f32[1,128,32,32]{3,2,1,0}, f32[128,128,3,3]{3,2,1,0}, f32[128]{0}), window={size=3x3 pad=1_1x1_1}, dim_labels=bf01_oi01->bf01, custom_call_target="__cudnn$convBiasActivationForward", backend_config={"cudnn_conv_backend_config":{"activation_mode":"kRelu","conv_result_scale":1,"leakyrelu_alpha":0,"side_input_scale":0},"force_earliest_schedule":false,"operation_queue_id":"0","wait_on_operation_queues":[]}
2025-03-07 02:43:38.793162: I external/local_xla/xla/service/gpu/autotuning/conv_algorithm_picker.cc:557] Omitted potentially buggy algorithm eng14{k25=2} for conv (f32[1,256,16,16]{3,2,1,0}, u8[0]{0}) custom-call(f32[1,128,16,16]{3,2,1,0}, f32[256,128,3,3]{3,2,1,0}, f32[256]{0}), window={size=3x3 pad=1_1x1_1}, dim_labels=bf01_oi01->bf01, custom_call_target="__cudnn$convBiasActivationForward", backend_config={"cudnn_conv_backend_config":{"activation_mode":"kRelu","conv_result_scale":1,"leakyrelu_alpha":0,"side_input_scale":0},"force_earliest_schedule":false,"operation_queue_id":"0","wait_on_operation_queues":[]}
2025-03-07 02:43:38.867057: I external/local_xla/xla/service/gpu/autotuning/conv_algorithm_picker.cc:557] Omitted potentially buggy algorithm eng14{k25=2} for conv (f32[1,256,16,16]{3,2,1,0}, u8[0]{0}) custom-call(f32[1,256,16,16]{3,2,1,0}, f32[256,256,3,3]{3,2,1,0}, f32[256]{0}), window={size=3x3 pad=1_1x1_1}, dim_labels=bf01_oi01->bf01, custom_call_target="__cudnn$convBiasActivationForward", backend_config={"cudnn_conv_backend_config":{"activation_mode":"kRelu","conv_result_scale":1,"leakyrelu_alpha":0,"side_input_scale":0},"force_earliest_schedule":false,"operation_queue_id":"0","wait_on_operation_queues":[]}
2025-03-07 02:43:38.965396: I external/local_xla/xla/service/gpu/autotuning/conv_algorithm_picker.cc:557] Omitted potentially buggy algorithm eng14{k25=2} for conv (f32[1,512,8,8]{3,2,1,0}, u8[0]{0}) custom-call(f32[1,256,8,8]{3,2,1,0}, f32[512,256,3,3]{3,2,1,0}, f32[512]{0}), window={size=3x3 pad=1_1x1_1}, dim_labels=bf01_oi01->bf01, custom_call_target="__cudnn$convBiasActivationForward", backend_config={"cudnn_conv_backend_config":{"activation_mode":"kRelu","conv_result_scale":1,"leakyrelu_alpha":0,"side_input_scale":0},"force_earliest_schedule":false,"operation_queue_id":"0","wait_on_operation_queues":[]}
2025-03-07 02:43:39.062206: I external/local_xla/xla/service/gpu/autotuning/conv_algorithm_picker.cc:557] Omitted potentially buggy algorithm eng14{k25=2} for conv (f32[1,512,8,8]{3,2,1,0}, u8[0]{0}) custom-call(f32[1,512,8,8]{3,2,1,0}, f32[512,512,3,3]{3,2,1,0}, f32[512]{0}), window={size=3x3 pad=1_1x1_1}, dim_labels=bf01_oi01->bf01, custom_call_target="__cudnn$convBiasActivationForward", backend_config={"cudnn_conv_backend_config":{"activation_mode":"kRelu","conv_result_scale":1,"leakyrelu_alpha":0,"side_input_scale":0},"force_earliest_schedule":false,"operation_queue_id":"0","wait_on_operation_queues":[]}
2025-03-07 02:43:39.205439: I external/local_xla/xla/service/gpu/autotuning/conv_algorithm_picker.cc:557] Omitted potentially buggy algorithm eng14{k25=2} for conv (f32[1,512,4,4]{3,2,1,0}, u8[0]{0}) custom-call(f32[1,512,4,4]{3,2,1,0}, f32[512,512,3,3]{3,2,1,0}, f32[512]{0}), window={size=3x3 pad=1_1x1_1}, dim_labels=bf01_oi01->bf01, custom_call_target="__cudnn$convBiasActivationForward", backend_config={"cudnn_conv_backend_config":{"activation_mode":"kRelu","conv_result_scale":1,"leakyrelu_alpha":0,"side_input_scale":0},"force_earliest_schedule":false,"operation_queue_id":"0","wait_on_operation_queues":[]}

101/101 ━━━━━━━━━━━━━━━━━━━━ 19s 132ms/step - binary_accuracy: 0.8126 - loss: 0.5205 - val_binary_accuracy: 0.9816 - val_loss: 0.0648
Epoch 2/5
101/101 ━━━━━━━━━━━━━━━━━━━━ 11s 110ms/step - binary_accuracy: 0.9894 - loss: 0.0417 - val_binary_accuracy: 0.9900 - val_loss: 0.0392
Epoch 3/5
101/101 ━━━━━━━━━━━━━━━━━━━━ 11s 110ms/step - binary_accuracy: 0.9927 - loss: 0.0254 - val_binary_accuracy: 0.9940 - val_loss: 0.0290
Epoch 4/5
101/101 ━━━━━━━━━━━━━━━━━━━━ 11s 110ms/step - binary_accuracy: 0.9952 - loss: 0.0176 - val_binary_accuracy: 0.9940 - val_loss: 0.0238
Epoch 5/5
101/101 ━━━━━━━━━━━━━━━━━━━━ 11s 111ms/step - binary_accuracy: 0.9963 - loss: 0.0136 - val_binary_accuracy: 0.9956 - val_loss: 0.0203

fig, (ax1, ax2) = plt.subplots(1, 2, figsize=(15, 5))

ax1.plot(history_vgg.history['binary_accuracy'])
ax1.plot(history_vgg.history['val_binary_accuracy'])
ax1.set_xticks(range(epochs))
ax1.set_xlabel("Epochs")
ax1.set_ylabel("Accuracy")
ax1.legend(["Training", "Validation"])

ax2.plot(history_vgg.history['loss'])
ax2.plot(history_vgg.history['val_loss'])
ax2.set_xticks(range(epochs))
ax2.set_xlabel("Epochs")
ax2.set_ylabel("Loss")
ax2.legend(["Training", "Validation"])

plt.show()

Evaluate the VGG16 model

images, labels = next(iter(ds_test))

fig, axs = plt.subplots(4, 4, figsize=(8, 8))
fig.suptitle('Random samples')

for i, ax in enumerate(axs.flatten()):
    prediction_proba = model_vgg.predict(tf.expand_dims(images[i], axis=0), verbose=0)
    ax.imshow(images[i] / 255)
    ax.set_title("Prediction: " + classes[int(prediction_proba.squeeze().round())])
    ax.axis("off")

plt.tight_layout()
plt.show()

_, y_test = zip(*ds_test.unbatch())
y_test = [y.numpy() for y in y_test]

prediction_proba = model_vgg.predict(ds_test, verbose=0)
y_pred = prediction_proba.squeeze().round()

print(classification_report(y_test, y_pred, digits=4))

ConfusionMatrixDisplay.from_predictions(y_test, y_pred, display_labels=["0 (negative)", "1 (positive)"])
plt.grid(False)
plt.show()

2025-03-07 02:44:31.798964: I tensorflow/core/framework/local_rendezvous.cc:405] Local rendezvous is aborting with status: OUT_OF_RANGE: End of sequence

              precision    recall  f1-score   support

           0     0.9984    0.9944    0.9964      1249
           1     0.9944    0.9984    0.9964      1251

    accuracy                         0.9964      2500
   macro avg     0.9964    0.9964    0.9964      2500
weighted avg     0.9964    0.9964    0.9964      2500