- GradientTape
- Module
- Tensor
- Variable
- keras
- Input
- layers
- Dense
- Flatten
- Conv1D
- Conv2D
- MaxPooling2D
- GlobalAveragePooling1D
- GlobalAveragePooling2D
- Dropout
- LSTM
- Bidirectional
- models
- load_model
- Sequential
- Model
- losses
- BinaryCrossentropy
- datasets
- fashion_mnist
- mnist
- optimizers
- Adam
- RMSprop
- callbacks
- Callback
- TensorBoard
- History
- applications
- InceptionV3
- MobileNetV2
- backend
- clear_session
- preprocessing
- image_dataset_from_directory
- image
- ImageDataGenerator
- DirectoryIterator
- load_img
- img_to_array
- text
- Tokenizer
- sequence
- pad_sequences
- utils
- to_categorical
- metrics
- mean_squared_error
- mean_absolute_error
- nn
- relu
- softmax
- optimizers
- Adam
- RMSprop
- data
- Dataset
- python
- ops
- math_ops
- squared_difference
- cast
- math_ops
- ops
- compat
- v1
- ConfigProto
- Session
- data
- get_output_shapes
- v2
- v1
再深的就因为OOM出不来了。。。
原图
卷积后输出
池化后输出