![](/img/trans.png)
[英]How to use keras image_dataset_from_directory with custom structures?
[英]How to use tfa.metrics.F1Score with image_dataset_from_directory correctly?
Colab 代碼在這里:
我正在關注此處的文檔以獲得多類預測的結果
當我訓練使用
#last layer
tf.keras.layers.Dense(2, activation='softmax')
model.compile(optimizer="adam",
loss=tf.keras.losses.CategoricalCrossentropy(),
metrics=[tf.keras.metrics.CategoricalAccuracy(),
tfa.metrics.F1Score(num_classes=2, average='macro')])
我明白了
144/144 [==] - 8s 54ms/step - loss: 0.0613 - categorical_accuracy: 0.9789 - f1_score: 0.9788 - val_loss: 0.0826 - val_categorical_accuracy: 0.9725 - val_f1_score: 0.9722
當我做:
model.evaluate(val_ds)
我明白了
16/16 [==] - 0s 15ms/step - loss: 0.0826 - categorical_accuracy: 0.9725 - f1_score: 0.9722
[0.08255868405103683, 0.9725490212440491, 0.9722140431404114]
我想在官方網站上使用metric.result
。 當我加載下面的代碼時,我得到0.4875028
這是錯誤的。 如何獲得正確的true_categories
predicted_categories
metric = tfa.metrics.F1Score(num_classes=2, average='macro')
predicted_categories = model.predict(val_ds)
true_categories = tf.concat([y for x, y in val_ds], axis=0).numpy()
metric.update_state(true_categories, predicted_categories)
result = metric.result()
print(result.numpy())
#0.4875028
這是我加載數據的方式
train_ds = tf.keras.preprocessing.image_dataset_from_directory(
main_folder,
validation_split=0.1,
subset="training",
label_mode='categorical',
seed=123,
image_size=(dim, dim))
val_ds = tf.keras.preprocessing.image_dataset_from_directory(
main_folder,
validation_split=0.1,
subset="validation",
label_mode='categorical',
seed=123,
image_size=(dim, dim))
來自: https://www.tensorflow.org/api_docs/python/tf/keras/preprocessing/image_dataset_from_directory
tf.keras.preprocessing.image_dataset_from_directory(
directory, labels='inferred', label_mode='int',
class_names=None, color_mode='rgb', batch_size=32, image_size=(256,
256), shuffle=True, seed=None, validation_split=None, subset=None,
interpolation='bilinear', follow_links=False
)
默認情況下, shuffle
是True
,這對您的val_ds
來說是個問題,我們不想隨機播放。
正確的指標是訓練期間報告的指標; 此外,我建議您也可以手動檢索驗證數據集並在對其進行預測后檢查指標(不一定通過flow_from_directory()
)。
聲明:本站的技術帖子網頁,遵循CC BY-SA 4.0協議,如果您需要轉載,請注明本站網址或者原文地址。任何問題請咨詢:yoyou2525@163.com.