reaarrange files
This commit is contained in:
@@ -0,0 +1,306 @@
|
||||
from unittest import skip
|
||||
import numpy as np
|
||||
import tensorflow as tf
|
||||
from tensorflow.keras.layers import (Conv2D, Conv2DTranspose, Dense, DepthwiseConv2D,
|
||||
GlobalAveragePooling2D, Layer,
|
||||
LayerNormalization, Flatten, Reshape, UpSampling2D)
|
||||
|
||||
def drop_path(inputs, drop_prob, is_training):
|
||||
# borrowed from https://github.com/rishigami/Swin-Transformer-TF/blob/main/swintransformer/model.py
|
||||
if (not is_training) or (drop_prob == 0.):
|
||||
return inputs
|
||||
|
||||
# Compute keep_prob
|
||||
keep_prob = 1.0 - drop_prob
|
||||
|
||||
# Compute drop_connect tensor
|
||||
random_tensor = keep_prob
|
||||
shape = (tf.shape(inputs)[0],) + (1,) * \
|
||||
(len(tf.shape(inputs)) - 1)
|
||||
random_tensor += tf.random.uniform(shape, dtype=inputs.dtype)
|
||||
binary_tensor = tf.floor(random_tensor)
|
||||
output = tf.math.divide(inputs, keep_prob) * binary_tensor
|
||||
return output
|
||||
|
||||
|
||||
class DropPath(tf.keras.layers.Layer):
|
||||
# borrowed from https://github.com/rishigami/Swin-Transformer-TF/blob/main/swintransformer/model.py
|
||||
def __init__(self, drop_prob=None):
|
||||
super().__init__()
|
||||
self.drop_prob = drop_prob
|
||||
|
||||
def call(self, x, training):
|
||||
return drop_path(x, self.drop_prob, training)
|
||||
|
||||
|
||||
class Block(Layer):
|
||||
r""" ConvNeXt Block. There are two equivalent implementations:
|
||||
(1) DwConv -> LayerNorm (channels_first) -> 1x1 Conv -> GELU -> 1x1 Conv; all in (N, C, H, W)
|
||||
(2) DwConv -> Permute to (N, H, W, C); LayerNorm (channels_last) -> Linear -> GELU -> Linear; Permute back
|
||||
We use (2) as we find it slightly faster in PyTorch
|
||||
|
||||
Args:
|
||||
dim (int): Number of input channels.
|
||||
drop_path (float): Stochastic depth rate. Default: 0.0
|
||||
layer_scale_init_value (float): Init value for Layer Scale. Default: 1e-6.
|
||||
"""
|
||||
|
||||
def __init__(self, dim, kernel_size=7, drop_path=0., layer_scale_init_value=1e-6, prefix=''):
|
||||
super().__init__()
|
||||
self.dwconv = DepthwiseConv2D(
|
||||
kernel_size=kernel_size, padding='same') # depthwise conv
|
||||
self.norm = LayerNormalization(epsilon=1e-6)
|
||||
# pointwise/1x1 convs, implemented with linear layers
|
||||
self.pwconv1 = Dense(4 * dim)
|
||||
self.act = tf.keras.activations.gelu
|
||||
self.pwconv2 = Dense(dim)
|
||||
self.drop_path = DropPath(drop_path)
|
||||
self.dim = dim
|
||||
self.layer_scale_init_value = layer_scale_init_value
|
||||
self.prefix = prefix
|
||||
|
||||
def build(self, input_shape):
|
||||
self.gamma = tf.Variable(
|
||||
initial_value=self.layer_scale_init_value * tf.ones((self.dim)),
|
||||
trainable=True,
|
||||
name=f'{self.prefix}/gamma')
|
||||
self.built = True
|
||||
|
||||
def call(self, x, training):
|
||||
input = x
|
||||
x = self.dwconv(x, training=training)
|
||||
# x = x.permute(0, 2, 3, 1) # (N, C, H, W) -> (N, H, W, C)
|
||||
x = self.norm(x, training=training)
|
||||
x = self.pwconv1(x, training=training)
|
||||
x = self.act(x)
|
||||
x = self.pwconv2(x, training=training)
|
||||
if self.gamma is not None:
|
||||
x = self.gamma * x
|
||||
# x = x.permute(0, 3, 1, 2) # (N, H, W, C) -> (N, C, H, W)
|
||||
|
||||
x = input + self.drop_path(x, training=training)
|
||||
return x
|
||||
|
||||
|
||||
class ConvNeXt(tf.keras.Model):
|
||||
r""" ConvNeXt
|
||||
A Tensorflow keras impl of : `A ConvNet for the 2020s` -
|
||||
https://arxiv.org/pdf/2201.03545.pdf
|
||||
|
||||
Args:
|
||||
num_classes (int): Number of classes for classification head. Default: 1000
|
||||
depths (tuple(int)): Number of blocks at each stage. Default: [3, 3, 9, 3]
|
||||
dims (int): Feature dimension at each stage. Default: [96, 192, 384, 768]
|
||||
include_top (bool): whether to add head or just use it as feature extractor. Default: True
|
||||
drop_path_rate (float): Stochastic depth rate. Default: 0.
|
||||
layer_scale_init_value (float): Init value for Layer Scale. Default: 1e-6.
|
||||
head_init_scale (float): Init scaling value for classifier weights and biases. Default: 1.
|
||||
"""
|
||||
|
||||
def __init__(self, num_classes=1000, kernel_size=7,
|
||||
depths=[3, 3, 9, 3], dims=[96, 192, 384, 768], include_top=True,
|
||||
drop_path_rate=0., layer_scale_init_value=1e-6, head_init_scale=1.,
|
||||
):
|
||||
super().__init__()
|
||||
self.include_top = include_top
|
||||
self.downsample_layers = [] # stem and 3 intermediate downsampling conv layers
|
||||
# stem = tf.keras.Sequential([
|
||||
# Conv2D(dims[0], kernel_size=4, strides=4, padding='same'),
|
||||
# LayerNormalization(epsilon=1e-6)]
|
||||
# )
|
||||
stem = [Conv2D(dims[0], kernel_size=4, strides=4, padding='same'), LayerNormalization(epsilon=1e-6)]
|
||||
self.downsample_layers.append(stem)
|
||||
for i in range(len(depths) - 1):
|
||||
# downsample_layer = tf.keras.Sequential([
|
||||
# LayerNormalization(epsilon=1e-6),
|
||||
# Conv2D(dims[i+1], kernel_size=2, strides=2, padding='same')]
|
||||
# )
|
||||
downsample_layer = [LayerNormalization(epsilon=1e-6), Conv2D(dims[i+1], kernel_size=2, strides=2, padding='same')]
|
||||
self.downsample_layers.append(downsample_layer)
|
||||
|
||||
self.stages = [] # 4 feature resolution stages, each consisting of multiple residual blocks
|
||||
dp_rates = [x for x in np.linspace(0, drop_path_rate, sum(depths))]
|
||||
cur = 0
|
||||
for i in range(len(depths)):
|
||||
stage = tf.keras.Sequential(
|
||||
[Block(dim=dims[i], drop_path=dp_rates[cur + j], kernel_size=kernel_size,
|
||||
layer_scale_init_value=layer_scale_init_value, prefix=f'block{i}') for j in range(depths[i])]
|
||||
)
|
||||
self.stages.append(stage)
|
||||
cur += depths[i]
|
||||
|
||||
if self.include_top:
|
||||
self.avg = GlobalAveragePooling2D()
|
||||
self.norm = LayerNormalization(epsilon=1e-6) # final norm layer
|
||||
self.head = Dense(num_classes)
|
||||
else:
|
||||
self.avg = None
|
||||
self.norm = None
|
||||
self.head = None
|
||||
|
||||
def forward_features(self, x, training):
|
||||
for i in range(len(self.downsample_layers)):
|
||||
# x = self.downsample_layers[i](x)
|
||||
x = self.downsample_layers[i][0](x, training=training)
|
||||
x = self.downsample_layers[i][1](x, training=training)
|
||||
x = self.stages[i](x, training=training)
|
||||
return x
|
||||
|
||||
def call(self, x, training):
|
||||
x = self.forward_features(x, training=training)
|
||||
return x
|
||||
|
||||
|
||||
class ConvNeXtReverse(tf.keras.Model):
|
||||
r""" ConvNeXt
|
||||
A Tensorflow keras impl of : `A ConvNet for the 2020s` -
|
||||
https://arxiv.org/pdf/2201.03545.pdf
|
||||
|
||||
Args:
|
||||
num_classes (int): Number of classes for classification head. Default: 1000
|
||||
depths (tuple(int)): Number of blocks at each stage. Default: [3, 3, 9, 3]
|
||||
dims (int): Feature dimension at each stage. Default: [96, 192, 384, 768]
|
||||
include_top (bool): whether to add head or just use it as feature extractor. Default: True
|
||||
drop_path_rate (float): Stochastic depth rate. Default: 0.
|
||||
layer_scale_init_value (float): Init value for Layer Scale. Default: 1e-6.
|
||||
head_init_scale (float): Init scaling value for classifier weights and biases. Default: 1.
|
||||
"""
|
||||
|
||||
def __init__(self,
|
||||
depths=[3, 3, 9, 3], dims=[96, 192, 384, 768], kernel_size=7,
|
||||
drop_path_rate=0., layer_scale_init_value=1e-6, head_init_scale=1.,
|
||||
):
|
||||
super().__init__()
|
||||
self.upsample_layers = [] # stem and 3 intermediate downsampling conv layers
|
||||
stem = tf.keras.Sequential([
|
||||
Conv2DTranspose(dims[0], kernel_size=4, strides=4, padding='same'),
|
||||
# UpSampling2D((4, 4), interpolation='nearest'),
|
||||
# Conv2D(dims[0], kernel_size=4, strides=1, padding='same'),
|
||||
# Dense(dims[0]),
|
||||
LayerNormalization(epsilon=1e-6)]
|
||||
)
|
||||
self.upsample_layers.append(stem)
|
||||
for i in range(len(depths) - 1):
|
||||
downsample_layer = tf.keras.Sequential([
|
||||
LayerNormalization(epsilon=1e-6),
|
||||
Conv2DTranspose(dims[i+1], kernel_size=2, strides=2, padding='same')
|
||||
# UpSampling2D((2,2), interpolation='nearest'),
|
||||
# Conv2D(dims[i+1], kernel_size=2, strides=1, padding='same'),
|
||||
# Dense(dims[i+1]),
|
||||
])
|
||||
self.upsample_layers.append(downsample_layer)
|
||||
|
||||
self.stages = [] # 4 feature resolution stages, each consisting of multiple residual blocks
|
||||
dp_rates = [x for x in np.linspace(0, drop_path_rate, sum(depths))]
|
||||
cur = 0
|
||||
for i in range(len(depths)):
|
||||
stage = tf.keras.Sequential(
|
||||
[Block(dim=dims[i], drop_path=dp_rates[cur + j], kernel_size=kernel_size,
|
||||
layer_scale_init_value=layer_scale_init_value, prefix=f'block{i}') for j in range(depths[i])]
|
||||
)
|
||||
self.stages.append(stage)
|
||||
cur += depths[i]
|
||||
|
||||
self.last = Dense(3, activation='tanh')
|
||||
|
||||
def call(self, x, training):
|
||||
for i in reversed(range(len(self.upsample_layers))):
|
||||
x = self.upsample_layers[i](x, training=training)
|
||||
x = self.stages[i](x, training=training)
|
||||
x = self.last(x, training=training)
|
||||
return x
|
||||
|
||||
|
||||
class ConvNeXtReverseLabels(tf.keras.Model):
|
||||
r""" ConvNeXt
|
||||
A Tensorflow keras impl of : `A ConvNet for the 2020s` -
|
||||
https://arxiv.org/pdf/2201.03545.pdf
|
||||
|
||||
Args:
|
||||
num_classes (int): Number of classes for classification head. Default: 1000
|
||||
depths (tuple(int)): Number of blocks at each stage. Default: [3, 3, 9, 3]
|
||||
dims (int): Feature dimension at each stage. Default: [96, 192, 384, 768]
|
||||
include_top (bool): whether to add head or just use it as feature extractor. Default: True
|
||||
drop_path_rate (float): Stochastic depth rate. Default: 0.
|
||||
layer_scale_init_value (float): Init value for Layer Scale. Default: 1e-6.
|
||||
head_init_scale (float): Init scaling value for classifier weights and biases. Default: 1.
|
||||
"""
|
||||
|
||||
def __init__(self,
|
||||
depths=[3, 3, 9, 3], dims=[96, 192, 384, 768],
|
||||
drop_path_rate=0., layer_scale_init_value=1e-6, head_init_scale=1.,
|
||||
nr_of_classes=7):
|
||||
super().__init__()
|
||||
self.downsample_layers = [] # stem and 3 intermediate downsampling conv layers
|
||||
stem = tf.keras.Sequential([
|
||||
Conv2DTranspose(dims[0], kernel_size=4, strides=4, padding='same'),
|
||||
LayerNormalization(epsilon=1e-6)]
|
||||
)
|
||||
self.downsample_layers.append(stem)
|
||||
for i in range(len(depths) - 1):
|
||||
downsample_layer = tf.keras.Sequential([
|
||||
LayerNormalization(epsilon=1e-6),
|
||||
Conv2DTranspose(dims[i+1], kernel_size=2, strides=2, padding='same')]
|
||||
)
|
||||
self.downsample_layers.append(downsample_layer)
|
||||
|
||||
self.stages = [] # 4 feature resolution stages, each consisting of multiple residual blocks
|
||||
dp_rates = [x for x in np.linspace(0, drop_path_rate, sum(depths))]
|
||||
cur = 0
|
||||
for i in range(len(depths)):
|
||||
stage = tf.keras.Sequential(
|
||||
[Block(dim=dims[i], drop_path=dp_rates[cur + j],
|
||||
layer_scale_init_value=layer_scale_init_value, prefix=f'block{i}') for j in range(depths[i])]
|
||||
)
|
||||
self.stages.append(stage)
|
||||
cur += depths[i]
|
||||
|
||||
# self.last = Dense(nr_of_classes, activation='softmax')
|
||||
self.last = Dense(nr_of_classes - 1)
|
||||
|
||||
def call(self, x, training):
|
||||
for i in reversed(range(len(self.downsample_layers))):
|
||||
x = self.downsample_layers[i](x, training=training)
|
||||
x = self.stages[i](x, training=training)
|
||||
x = self.last(x, training=training)
|
||||
return x
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
model_urls = {
|
||||
"convnext_tiny_224": "https://github.com/bamps53/convnext-tf/releases/download/v0.1/convnext_tiny_1k_224_ema.h5",
|
||||
"convnext_small_224": "https://github.com/bamps53/convnext-tf/releases/download/v0.1/convnext_small_1k_224_ema.h5",
|
||||
"convnext_base_224": "https://github.com/bamps53/convnext-tf/releases/download/v0.1/convnext_base_22k_1k_224.h5",
|
||||
"convnext_base_384": "https://github.com/bamps53/convnext-tf/releases/download/v0.1/convnext_base_22k_1k_384.h5",
|
||||
"convnext_large_224": "https://github.com/bamps53/convnext-tf/releases/download/v0.1/convnext_large_22k_1k_224.h5",
|
||||
"convnext_large_384": "https://github.com/bamps53/convnext-tf/releases/download/v0.1/convnext_large_22k_1k_384.h5",
|
||||
"convnext_xlarge_224": "https://github.com/bamps53/convnext-tf/releases/download/v0.1/convnext_xlarge_22k_1k_224_ema.h5",
|
||||
"convnext_xlarge_384": "https://github.com/bamps53/convnext-tf/releases/download/v0.1/convnext_xlarge_22k_1k_384_ema.h5",
|
||||
}
|
||||
|
||||
|
||||
model_configs = dict(
|
||||
convnext_tiny=dict(
|
||||
depths=[3, 3, 9, 3],
|
||||
dims=[8, 16, 32, 512]
|
||||
),
|
||||
convnext_small=dict(
|
||||
depths=[3, 3, 27, 3],
|
||||
dims=[96, 192, 384, 768]
|
||||
),
|
||||
convnext_base=dict(
|
||||
depths=[3, 3, 27, 3],
|
||||
dims=[128, 256, 512, 1024]
|
||||
),
|
||||
convnext_large=dict(
|
||||
depths=[3, 3, 27, 3],
|
||||
dims=[192, 384, 768, 1536]
|
||||
),
|
||||
convnext_xlarge=dict(
|
||||
depths=[3, 3, 27, 3],
|
||||
dims=[256, 512, 1024, 2048]
|
||||
),
|
||||
)
|
||||
|
||||
Reference in New Issue
Block a user