时间戳

首页 > 解决方案 > tf.kerasConv2D 中的 InvalidArgumentError

tensorflow - tf.kerasConv2D 中的 InvalidArgumentError

问题描述

我做了一个卷积网络，在某些层有膨胀，训练时出现以下错误： tensorflow.python.framework.errors_impl.InvalidArgumentError: padded_shape[0]=170 is not divisible by block_shape[0]=4

错误发生在名为 encoder_5 的层上，第一次应用膨胀。当我在该层中使用时不会出现错误padding="valid"，但这不是一种选择，因为我需要保持尺寸以便能够在以后的阶段进行连接。我不明白这个错误来自哪里，填充似乎无法正常工作。错误与此处相同：https://github.com/tensorflow/tensorflow/issues/28788，但是我可以毫无问题地运行错误报告中的代码，因此我怀疑这是tf.keras中的错误。我究竟做错了什么？

我正在使用张量流 2.2.0。

完整的堆栈跟踪：

(4, 160, 90, 4096)
(4, 160, 90, 4096)
2020-08-14 16:01:25.577940: I tensorflow/core/kernels/data/shuffle_dataset_op.cc:143] Filling up shuffle buffer (this may take a while): 1 of 4
(4, 160, 90, 4096)
2020-08-14 16:02:01.700859: I tensorflow/core/kernels/data/shuffle_dataset_op.cc:143] Filling up shuffle buffer (this may take a while): 2 of 4
(4, 160, 90, 4096)
2020-08-14 16:02:38.701768: I tensorflow/core/kernels/data/shuffle_dataset_op.cc:143] Filling up shuffle buffer (this may take a while): 3 of 4
(2, 160, 90, 4096)
2020-08-14 16:02:46.407833: I tensorflow/core/kernels/data/shuffle_dataset_op.cc:193] Shuffle buffer filled.
(2, 160, 90, 4096)
(2, 160, 90, 4096)
Train for 4 steps, validate for 1 steps
Epoch 1/500
(4, 160, 90, 4096)
2020-08-14 16:03:42.476405: I tensorflow/core/kernels/data/shuffle_dataset_op.cc:143] Filling up shuffle buffer (this may take a while): 1 of 4
(4, 160, 90, 4096)
2020-08-14 16:04:18.252608: I tensorflow/core/kernels/data/shuffle_dataset_op.cc:143] Filling up shuffle buffer (this may take a while): 2 of 4
(4, 160, 90, 4096)
2020-08-14 16:04:52.134605: I tensorflow/core/kernels/data/shuffle_dataset_op.cc:143] Filling up shuffle buffer (this may take a while): 3 of 4
(2, 160, 90, 4096)
2020-08-14 16:05:11.404984: I tensorflow/core/kernels/data/shuffle_dataset_op.cc:193] Shuffle buffer filled.
2020-08-14 16:17:53.293766: W tensorflow/core/framework/op_kernel.cc:1622] OP_REQUIRES failed at spacetobatch_op.cc:219 : Invalid argument: padded_shape[0]=170 is not divisible by block_shape[0]=4
2020-08-14 16:17:53.302512: W tensorflow/core/common_runtime/base_collective_executor.cc:216] BaseCollectiveExecutor::StartAbort Invalid argument: padded_shape[0]=170 is not divisible by block_shape[0]=4
     [[{{node cutie/conv2d_5/SpaceToBatchND}}]]
1/4 [======>.......................] - ETA: 44:43WARNING:tensorflow:Early stopping conditioned on metric `val_loss` which is not available. Available metrics are: 
2020-08-14 16:17:54 WARNING  tensorflow Early stopping conditioned on metric `val_loss` which is not available. Available metrics are: 
Traceback (most recent call last):
  File "/Users/aleksandra/Projects/NLP/nlp-entity-extraction/scripts/train_locally.py", line 53, in <module>
    train_and_save_model(training_params, model_meta)
  File "/Users/aleksandra/Projects/NLP/nlp-entity-extraction/scripts/train_locally.py", line 33, in train_and_save_model
    train_and_save_grid_model(pipeline_folder)
  File "/Users/aleksandra/Projects/NLP/nlp-entity-extraction/train_pipeline/step_3_model_training/train_cutie_model.py", line 42, in train_and_save_grid_model
    test_data_gen = train_model(model, training_params, model_meta)
  File "/Users/aleksandra/Projects/NLP/nlp-entity-extraction/training/training_utils.py", line 45, in train_model
    verbose=True)
  File "/Users/aleksandra/Projects/NLP/nlp-entity-extraction/venv/lib/python3.7/site-packages/tensorflow_core/python/keras/engine/training.py", line 728, in fit
    use_multiprocessing=use_multiprocessing)
  File "/Users/aleksandra/Projects/NLP/nlp-entity-extraction/venv/lib/python3.7/site-packages/tensorflow_core/python/keras/engine/training_v2.py", line 324, in fit
    total_epochs=epochs)
  File "/Users/aleksandra/Projects/NLP/nlp-entity-extraction/venv/lib/python3.7/site-packages/tensorflow_core/python/keras/engine/training_v2.py", line 123, in run_one_epoch
    batch_outs = execution_function(iterator)
  File "/Users/aleksandra/Projects/NLP/nlp-entity-extraction/venv/lib/python3.7/site-packages/tensorflow_core/python/keras/engine/training_v2_utils.py", line 86, in execution_function
    distributed_function(input_fn))
  File "/Users/aleksandra/Projects/NLP/nlp-entity-extraction/venv/lib/python3.7/site-packages/tensorflow_core/python/eager/def_function.py", line 457, in __call__
    result = self._call(*args, **kwds)
  File "/Users/aleksandra/Projects/NLP/nlp-entity-extraction/venv/lib/python3.7/site-packages/tensorflow_core/python/eager/def_function.py", line 520, in _call
    return self._stateless_fn(*args, **kwds)
  File "/Users/aleksandra/Projects/NLP/nlp-entity-extraction/venv/lib/python3.7/site-packages/tensorflow_core/python/eager/function.py", line 1823, in __call__
    return graph_function._filtered_call(args, kwargs)  # pylint: disable=protected-access
  File "/Users/aleksandra/Projects/NLP/nlp-entity-extraction/venv/lib/python3.7/site-packages/tensorflow_core/python/eager/function.py", line 1141, in _filtered_call
    self.captured_inputs)
  File "/Users/aleksandra/Projects/NLP/nlp-entity-extraction/venv/lib/python3.7/site-packages/tensorflow_core/python/eager/function.py", line 1224, in _call_flat
    ctx, args, cancellation_manager=cancellation_manager)
  File "/Users/aleksandra/Projects/NLP/nlp-entity-extraction/venv/lib/python3.7/site-packages/tensorflow_core/python/eager/function.py", line 511, in call
    ctx=ctx)
  File "/Users/aleksandra/Projects/NLP/nlp-entity-extraction/venv/lib/python3.7/site-packages/tensorflow_core/python/eager/execute.py", line 67, in quick_execute
    six.raise_from(core._status_to_exception(e.code, message), None)
  File "<string>", line 3, in raise_from
tensorflow.python.framework.errors_impl.InvalidArgumentError:  padded_shape[0]=170 is not divisible by block_shape[0]=4
     [[node cutie/conv2d_5/SpaceToBatchND (defined at /venv/lib/python3.7/site-packages/tensorflow_core/python/framework/ops.py:1751) ]] [Op:__inference_distributed_function_2737]

模型：

import tensorflow as tf

def model(labels):
    embeddings = tf.keras.Input(
        shape=(160, 90, 4096),
        dtype=tf.float32,
        name="embedding_grid",
    )

    # encoder
    n_filters = 4096 // 2

    encoder_1 = tf.keras.layers.Conv2D(
        filters=n_filters,
        kernel_size=[3, 5],
        padding="same",
        use_bias=True,
        activation="relu",
    )(embeddings)
    encoder_2 = tf.keras.layers.Conv2D(
        filters=n_filters,
        kernel_size=[3, 5],
        padding="same",
        use_bias=True,
        activation="relu",
    )(encoder_1)
    encoder_3 = tf.keras.layers.Conv2D(
        filters=n_filters,
        kernel_size=[3, 5],
        padding="same",
        use_bias=True,
        activation="relu",
    )(encoder_2)
    encoder_4 = tf.keras.layers.Conv2D(
        filters=n_filters,
        kernel_size=[3, 5],
        padding="same",
        use_bias=True,
        activation="relu",
    )(encoder_3)
    encoder_5 = tf.keras.layers.Conv2D(
        filters=n_filters,
        kernel_size=[3, 5],
        padding="same",
        use_bias=True,
        activation="relu",
        dilation_rate=(2, 2),
    )(encoder_4)
    encoder_6 = tf.keras.layers.Conv2D(
        filters=n_filters,
        kernel_size=[3, 5],
        padding="same",
        use_bias=True,
        activation="relu",
        dilation_rate=(4, 4),
    )(encoder_5)
    encoder_7 = tf.keras.layers.Conv2D(
        filters=n_filters,
        kernel_size=[3, 5],
        padding="same",
        use_bias=True,
        activation="relu",
        dilation_rate=(8, 8),
    )(encoder_6)
    encoder_8 = tf.keras.layers.Conv2D(
        filters=n_filters,
        kernel_size=[3, 5],
        padding="same",
        use_bias=True,
        activation="relu",
        dilation_rate=(16, 16),
    )(encoder_7)

    # Atrous Spatial Pyramid Pooling module

    aspp_1 = tf.keras.layers.Conv2D(
        filters=n_filters,
        kernel_size=[3, 5],
        padding="same",
        use_bias=True,
        activation="relu",
        dilation_rate=4,
    )(encoder_8)

    aspp_2 = tf.keras.layers.Conv2D(
        filters=n_filters,
        kernel_size=[3, 5],
        padding="same",
        use_bias=True,
        activation="relu",
        dilation_rate=8,
    )(encoder_8)

    aspp_3 = tf.keras.layers.Conv2D(
        filters=n_filters,
        kernel_size=[3, 5],
        padding="same",
        use_bias=True,
        activation="relu",
        dilation_rate=16,
    )(encoder_8)

    reduced = tf.reduce_mean(aspp_3, [1, 2], keepdims=True)
    global_pool = tf.image.resize(reduced, [tf.shape(aspp_3)[1], tf.shape(aspp_3)[2]], method='nearest',
                                  name="global_pool")

    # global_pool = tf.keras.layers.GlobalMaxPool2D()(encoder_8)
    aspp_concat = tf.concat([aspp_1, aspp_2, aspp_3, global_pool], axis=3)
    aspp_1x1 = tf.keras.layers.Conv2D(
        filters=n_filters,
        kernel_size=[1, 1],
        padding="same",
        use_bias=True,
        activation="relu",
    )(aspp_concat)

    # combine low level features
    concat = tf.concat([encoder_1, aspp_1x1], axis=3)
    decoder = tf.keras.layers.Conv2D(
        filters=64,
        kernel_size=[3, 5],
        padding="same",
        use_bias=True,
        activation="relu",
    )(concat)

    # classification
    logits = tf.keras.layers.Conv2D(
        filters=len(labels),
        kernel_size=[1, 1],
        padding="same",
        use_bias=True,
        activation="relu",
        name="logits",
    )(decoder)
    softmax = tf.keras.layers.Softmax(axis=3, name="softmax")(logits)
    pred_ids = tf.argmax(softmax, axis=2, output_type=tf.int32)

    loss = {"logits": "categorical_crossentropy"}

    model = tf.keras.Model(inputs=[embeddings], outputs=[logits, pred_ids], name="cutie")
    model.compile(
        optimizer=tf.keras.optimizers.Adam(
            learning_rate=0.0001, decay=0.0001 / 500
        ),
        loss=loss,
        metrics={"logits": [tf.keras.metrics.CategoricalAccuracy()]},
    )

标签： tensorflowkerasdeep-learningconv-neural-network

解决方案

事实证明，输入数据并非都具有相同的维度......

推荐阅读