paddlex 2.0.0rc4__py3-none-any.whl → 3.0.0b2__py3-none-any.whl

paddlex/ppcls/modeling/architectures/regnet.py

@@ -1,383 +0,0 @@
-# copyright (c) 2020 PaddlePaddle Authors. All Rights Reserve.
-#
-# Licensed under the Apache License, Version 2.0 (the "License");
-# you may not use this file except in compliance with the License.
-# You may obtain a copy of the License at
-#
-#    http://www.apache.org/licenses/LICENSE-2.0
-#
-# Unless required by applicable law or agreed to in writing, software
-# distributed under the License is distributed on an "AS IS" BASIS,
-# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-# See the License for the specific language governing permissions and
-# limitations under the License.
-
-from __future__ import absolute_import
-from __future__ import division
-from __future__ import print_function
-
-import numpy as np
-import paddle
-from paddle import ParamAttr
-import paddle.nn as nn
-import paddle.nn.functional as F
-from paddle.nn import Conv2D, BatchNorm, Linear, Dropout
-from paddle.nn import AdaptiveAvgPool2D, MaxPool2D, AvgPool2D
-from paddle.nn.initializer import Uniform
-import math
-
-__all__ = [
-    "RegNetX_200MF", "RegNetX_4GF", "RegNetX_32GF", "RegNetY_200MF",
-    "RegNetY_4GF", "RegNetY_32GF"
-]
-
-
-def quantize_float(f, q):
-    """Converts a float to closest non-zero int divisible by q."""
-    return int(round(f / q) * q)
-
-
-def adjust_ws_gs_comp(ws, bms, gs):
-    """Adjusts the compatibility of widths and groups."""
-    ws_bot = [int(w * b) for w, b in zip(ws, bms)]
-    gs = [min(g, w_bot) for g, w_bot in zip(gs, ws_bot)]
-    ws_bot = [quantize_float(w_bot, g) for w_bot, g in zip(ws_bot, gs)]
-    ws = [int(w_bot / b) for w_bot, b in zip(ws_bot, bms)]
-    return ws, gs
-
-
-def get_stages_from_blocks(ws, rs):
-    """Gets ws/ds of network at each stage from per block values."""
-    ts = [
-        w != wp or r != rp
-        for w, wp, r, rp in zip(ws + [0], [0] + ws, rs + [0], [0] + rs)
-    ]
-    s_ws = [w for w, t in zip(ws, ts[:-1]) if t]
-    s_ds = np.diff([d for d, t in zip(range(len(ts)), ts) if t]).tolist()
-    return s_ws, s_ds
-
-
-def generate_regnet(w_a, w_0, w_m, d, q=8):
-    """Generates per block ws from RegNet parameters."""
-    assert w_a >= 0 and w_0 > 0 and w_m > 1 and w_0 % q == 0
-    ws_cont = np.arange(d) * w_a + w_0
-    ks = np.round(np.log(ws_cont / w_0) / np.log(w_m))
-    ws = w_0 * np.power(w_m, ks)
-    ws = np.round(np.divide(ws, q)) * q
-    num_stages, max_stage = len(np.unique(ws)), ks.max() + 1
-    ws, ws_cont = ws.astype(int).tolist(), ws_cont.tolist()
-    return ws, num_stages, max_stage, ws_cont
-
-
-class ConvBNLayer(nn.Layer):
-    def __init__(self,
-                 num_channels,
-                 num_filters,
-                 filter_size,
-                 stride=1,
-                 groups=1,
-                 padding=0,
-                 act=None,
-                 name=None):
-        super(ConvBNLayer, self).__init__()
-
-        self._conv = Conv2D(
-            in_channels=num_channels,
-            out_channels=num_filters,
-            kernel_size=filter_size,
-            stride=stride,
-            padding=padding,
-            groups=groups,
-            weight_attr=ParamAttr(name=name + ".conv2d.output.1.w_0"),
-            bias_attr=ParamAttr(name=name + ".conv2d.output.1.b_0"))
-        bn_name = name + "_bn"
-        self._batch_norm = BatchNorm(
-            num_filters,
-            act=act,
-            param_attr=ParamAttr(name=bn_name + ".output.1.w_0"),
-            bias_attr=ParamAttr(bn_name + ".output.1.b_0"),
-            moving_mean_name=bn_name + "_mean",
-            moving_variance_name=bn_name + "_variance")
-
-    def forward(self, inputs):
-        y = self._conv(inputs)
-        y = self._batch_norm(y)
-        return y
-
-
-class BottleneckBlock(nn.Layer):
-    def __init__(self,
-                 num_channels,
-                 num_filters,
-                 stride,
-                 bm,
-                 gw,
-                 se_on,
-                 se_r,
-                 shortcut=True,
-                 name=None):
-        super(BottleneckBlock, self).__init__()
-
-        # Compute the bottleneck width
-        w_b = int(round(num_filters * bm))
-        # Compute the number of groups
-        num_gs = w_b // gw
-        self.se_on = se_on
-        self.conv0 = ConvBNLayer(
-            num_channels=num_channels,
-            num_filters=w_b,
-            filter_size=1,
-            padding=0,
-            act="relu",
-            name=name + "_branch2a")
-        self.conv1 = ConvBNLayer(
-            num_channels=w_b,
-            num_filters=w_b,
-            filter_size=3,
-            stride=stride,
-            padding=1,
-            groups=num_gs,
-            act="relu",
-            name=name + "_branch2b")
-        if se_on:
-            w_se = int(round(num_channels * se_r))
-            self.se_block = SELayer(
-                num_channels=w_b,
-                num_filters=w_b,
-                reduction_ratio=w_se,
-                name=name + "_branch2se")
-        self.conv2 = ConvBNLayer(
-            num_channels=w_b,
-            num_filters=num_filters,
-            filter_size=1,
-            act=None,
-            name=name + "_branch2c")
-
-        if not shortcut:
-            self.short = ConvBNLayer(
-                num_channels=num_channels,
-                num_filters=num_filters,
-                filter_size=1,
-                stride=stride,
-                name=name + "_branch1")
-
-        self.shortcut = shortcut
-
-    def forward(self, inputs):
-        y = self.conv0(inputs)
-        conv1 = self.conv1(y)
-        if self.se_on:
-            conv1 = self.se_block(conv1)
-        conv2 = self.conv2(conv1)
-
-        if self.shortcut:
-            short = inputs
-        else:
-            short = self.short(inputs)
-
-        y = paddle.add(x=short, y=conv2)
-        y = F.relu(y)
-        return y
-
-
-class SELayer(nn.Layer):
-    def __init__(self, num_channels, num_filters, reduction_ratio, name=None):
-        super(SELayer, self).__init__()
-
-        self.pool2d_gap = AdaptiveAvgPool2D(1)
-
-        self._num_channels = num_channels
-
-        med_ch = int(num_channels / reduction_ratio)
-        stdv = 1.0 / math.sqrt(num_channels * 1.0)
-        self.squeeze = Linear(
-            num_channels,
-            med_ch,
-            weight_attr=ParamAttr(
-                initializer=Uniform(-stdv, stdv), name=name + "_sqz_weights"),
-            bias_attr=ParamAttr(name=name + "_sqz_offset"))
-
-        stdv = 1.0 / math.sqrt(med_ch * 1.0)
-        self.excitation = Linear(
-            med_ch,
-            num_filters,
-            weight_attr=ParamAttr(
-                initializer=Uniform(-stdv, stdv), name=name + "_exc_weights"),
-            bias_attr=ParamAttr(name=name + "_exc_offset"))
-
-    def forward(self, input):
-        pool = self.pool2d_gap(input)
-        pool = paddle.reshape(pool, shape=[-1, self._num_channels])
-        squeeze = self.squeeze(pool)
-        squeeze = F.relu(squeeze)
-        excitation = self.excitation(squeeze)
-        excitation = F.sigmoid(excitation)
-        excitation = paddle.reshape(
-            excitation, shape=[-1, self._num_channels, 1, 1])
-        out = input * excitation
-        return out
-
-
-class RegNet(nn.Layer):
-    def __init__(self,
-                 w_a,
-                 w_0,
-                 w_m,
-                 d,
-                 group_w,
-                 bot_mul,
-                 q=8,
-                 se_on=False,
-                 class_dim=1000):
-        super(RegNet, self).__init__()
-
-        # Generate RegNet ws per block
-        b_ws, num_s, max_s, ws_cont = generate_regnet(w_a, w_0, w_m, d, q)
-        # Convert to per stage format
-        ws, ds = get_stages_from_blocks(b_ws, b_ws)
-        # Generate group widths and bot muls
-        gws = [group_w for _ in range(num_s)]
-        bms = [bot_mul for _ in range(num_s)]
-        # Adjust the compatibility of ws and gws
-        ws, gws = adjust_ws_gs_comp(ws, bms, gws)
-        # Use the same stride for each stage
-        ss = [2 for _ in range(num_s)]
-        # Use SE for RegNetY
-        se_r = 0.25
-        # Construct the model
-        # Group params by stage
-        stage_params = list(zip(ds, ws, ss, bms, gws))
-        # Construct the stem
-        stem_type = "simple_stem_in"
-        stem_w = 32
-        block_type = "res_bottleneck_block"
-
-        self.conv = ConvBNLayer(
-            num_channels=3,
-            num_filters=stem_w,
-            filter_size=3,
-            stride=2,
-            padding=1,
-            act="relu",
-            name="stem_conv")
-
-        self.block_list = []
-        for block, (d, w_out, stride, bm, gw) in enumerate(stage_params):
-            shortcut = False
-            for i in range(d):
-                num_channels = stem_w if block == i == 0 else in_channels
-                # Stride apply to the first block of the stage
-                b_stride = stride if i == 0 else 1
-                conv_name = "s" + str(block + 1) + "_b" + str(i +
-                                                              1)  # chr(97 + i)
-                bottleneck_block = self.add_sublayer(
-                    conv_name,
-                    BottleneckBlock(
-                        num_channels=num_channels,
-                        num_filters=w_out,
-                        stride=b_stride,
-                        bm=bm,
-                        gw=gw,
-                        se_on=se_on,
-                        se_r=se_r,
-                        shortcut=shortcut,
-                        name=conv_name))
-                in_channels = w_out
-                self.block_list.append(bottleneck_block)
-                shortcut = True
-
-        self.pool2d_avg = AdaptiveAvgPool2D(1)
-
-        self.pool2d_avg_channels = w_out
-
-        stdv = 1.0 / math.sqrt(self.pool2d_avg_channels * 1.0)
-
-        self.out = Linear(
-            self.pool2d_avg_channels,
-            class_dim,
-            weight_attr=ParamAttr(
-                initializer=Uniform(-stdv, stdv), name="fc_0.w_0"),
-            bias_attr=ParamAttr(name="fc_0.b_0"))
-
-    def forward(self, inputs):
-        y = self.conv(inputs)
-        for block in self.block_list:
-            y = block(y)
-        y = self.pool2d_avg(y)
-        y = paddle.reshape(y, shape=[-1, self.pool2d_avg_channels])
-        y = self.out(y)
-        return y
-
-
-def RegNetX_200MF(**args):
-    model = RegNet(
-        w_a=36.44, w_0=24, w_m=2.49, d=13, group_w=8, bot_mul=1.0, q=8, **args)
-    return model
-
-
-def RegNetX_4GF(**args):
-    model = RegNet(
-        w_a=38.65,
-        w_0=96,
-        w_m=2.43,
-        d=23,
-        group_w=40,
-        bot_mul=1.0,
-        q=8,
-        **args)
-    return model
-
-
-def RegNetX_32GF(**args):
-    model = RegNet(
-        w_a=69.86,
-        w_0=320,
-        w_m=2.0,
-        d=23,
-        group_w=168,
-        bot_mul=1.0,
-        q=8,
-        **args)
-    return model
-
-
-def RegNetY_200MF(**args):
-    model = RegNet(
-        w_a=36.44,
-        w_0=24,
-        w_m=2.49,
-        d=13,
-        group_w=8,
-        bot_mul=1.0,
-        q=8,
-        se_on=True,
-        **args)
-    return model
-
-
-def RegNetY_4GF(**args):
-    model = RegNet(
-        w_a=31.41,
-        w_0=96,
-        w_m=2.24,
-        d=22,
-        group_w=64,
-        bot_mul=1.0,
-        q=8,
-        se_on=True,
-        **args)
-    return model
-
-
-def RegNetY_32GF(**args):
-    model = RegNet(
-        w_a=115.89,
-        w_0=232,
-        w_m=2.53,
-        d=20,
-        group_w=232,
-        bot_mul=1.0,
-        q=8,
-        se_on=True,
-        **args)
-    return model

paddlex/ppcls/modeling/architectures/repvgg.py

@@ -1,339 +0,0 @@
-import paddle.nn as nn
-import paddle
-import numpy as np
-
-__all__ = [
-    'RepVGG',
-    'RepVGG_A0',
-    'RepVGG_A1',
-    'RepVGG_A2',
-    'RepVGG_B0',
-    'RepVGG_B1',
-    'RepVGG_B2',
-    'RepVGG_B3',
-    'RepVGG_B1g2',
-    'RepVGG_B1g4',
-    'RepVGG_B2g2',
-    'RepVGG_B2g4',
-    'RepVGG_B3g2',
-    'RepVGG_B3g4',
-]
-
-
-class ConvBN(nn.Layer):
-    def __init__(self,
-                 in_channels,
-                 out_channels,
-                 kernel_size,
-                 stride,
-                 padding,
-                 groups=1):
-        super(ConvBN, self).__init__()
-        self.conv = nn.Conv2D(
-            in_channels=in_channels,
-            out_channels=out_channels,
-            kernel_size=kernel_size,
-            stride=stride,
-            padding=padding,
-            groups=groups,
-            bias_attr=False)
-        self.bn = nn.BatchNorm2D(num_features=out_channels)
-
-    def forward(self, x):
-        y = self.conv(x)
-        y = self.bn(y)
-        return y
-
-
-class RepVGGBlock(nn.Layer):
-    def __init__(self,
-                 in_channels,
-                 out_channels,
-                 kernel_size,
-                 stride=1,
-                 padding=0,
-                 dilation=1,
-                 groups=1,
-                 padding_mode='zeros'):
-        super(RepVGGBlock, self).__init__()
-        self.in_channels = in_channels
-        self.out_channels = out_channels
-        self.kernel_size = kernel_size
-        self.stride = stride
-        self.padding = padding
-        self.dilation = dilation
-        self.groups = groups
-        self.padding_mode = padding_mode
-
-        assert kernel_size == 3
-        assert padding == 1
-
-        padding_11 = padding - kernel_size // 2
-
-        self.nonlinearity = nn.ReLU()
-
-        self.rbr_identity = nn.BatchNorm2D(
-            num_features=in_channels
-        ) if out_channels == in_channels and stride == 1 else None
-        self.rbr_dense = ConvBN(
-            in_channels=in_channels,
-            out_channels=out_channels,
-            kernel_size=kernel_size,
-            stride=stride,
-            padding=padding,
-            groups=groups)
-        self.rbr_1x1 = ConvBN(
-            in_channels=in_channels,
-            out_channels=out_channels,
-            kernel_size=1,
-            stride=stride,
-            padding=padding_11,
-            groups=groups)
-
-    def forward(self, inputs):
-        if not self.training:
-            return self.nonlinearity(self.rbr_reparam(inputs))
-
-        if self.rbr_identity is None:
-            id_out = 0
-        else:
-            id_out = self.rbr_identity(inputs)
-        return self.nonlinearity(
-            self.rbr_dense(inputs) + self.rbr_1x1(inputs) + id_out)
-
-    def eval(self):
-        if not hasattr(self, 'rbr_reparam'):
-            self.rbr_reparam = nn.Conv2D(
-                in_channels=self.in_channels,
-                out_channels=self.out_channels,
-                kernel_size=self.kernel_size,
-                stride=self.stride,
-                padding=self.padding,
-                dilation=self.dilation,
-                groups=self.groups,
-                padding_mode=self.padding_mode)
-        self.training = False
-        kernel, bias = self.get_equivalent_kernel_bias()
-        self.rbr_reparam.weight.set_value(kernel)
-        self.rbr_reparam.bias.set_value(bias)
-        for layer in self.sublayers():
-            layer.eval()
-
-    def get_equivalent_kernel_bias(self):
-        kernel3x3, bias3x3 = self._fuse_bn_tensor(self.rbr_dense)
-        kernel1x1, bias1x1 = self._fuse_bn_tensor(self.rbr_1x1)
-        kernelid, biasid = self._fuse_bn_tensor(self.rbr_identity)
-        return kernel3x3 + self._pad_1x1_to_3x3_tensor(
-            kernel1x1) + kernelid, bias3x3 + bias1x1 + biasid
-
-    def _pad_1x1_to_3x3_tensor(self, kernel1x1):
-        if kernel1x1 is None:
-            return 0
-        else:
-            return nn.functional.pad(kernel1x1, [1, 1, 1, 1])
-
-    def _fuse_bn_tensor(self, branch):
-        if branch is None:
-            return 0, 0
-        if isinstance(branch, ConvBN):
-            kernel = branch.conv.weight
-            running_mean = branch.bn._mean
-            running_var = branch.bn._variance
-            gamma = branch.bn.weight
-            beta = branch.bn.bias
-            eps = branch.bn._epsilon
-        else:
-            assert isinstance(branch, nn.BatchNorm2D)
-            if not hasattr(self, 'id_tensor'):
-                input_dim = self.in_channels // self.groups
-                kernel_value = np.zeros(
-                    (self.in_channels, input_dim, 3, 3), dtype=np.float32)
-                for i in range(self.in_channels):
-                    kernel_value[i, i % input_dim, 1, 1] = 1
-                self.id_tensor = paddle.to_tensor(kernel_value)
-            kernel = self.id_tensor
-            running_mean = branch._mean
-            running_var = branch._variance
-            gamma = branch.weight
-            beta = branch.bias
-            eps = branch._epsilon
-        std = (running_var + eps).sqrt()
-        t = (gamma / std).reshape((-1, 1, 1, 1))
-        return kernel * t, beta - running_mean * gamma / std
-
-
-class RepVGG(nn.Layer):
-    def __init__(self,
-                 num_blocks,
-                 width_multiplier=None,
-                 override_groups_map=None,
-                 class_dim=1000):
-        super(RepVGG, self).__init__()
-
-        assert len(width_multiplier) == 4
-        self.override_groups_map = override_groups_map or dict()
-
-        assert 0 not in self.override_groups_map
-
-        self.in_planes = min(64, int(64 * width_multiplier[0]))
-
-        self.stage0 = RepVGGBlock(
-            in_channels=3,
-            out_channels=self.in_planes,
-            kernel_size=3,
-            stride=2,
-            padding=1)
-        self.cur_layer_idx = 1
-        self.stage1 = self._make_stage(
-            int(64 * width_multiplier[0]), num_blocks[0], stride=2)
-        self.stage2 = self._make_stage(
-            int(128 * width_multiplier[1]), num_blocks[1], stride=2)
-        self.stage3 = self._make_stage(
-            int(256 * width_multiplier[2]), num_blocks[2], stride=2)
-        self.stage4 = self._make_stage(
-            int(512 * width_multiplier[3]), num_blocks[3], stride=2)
-        self.gap = nn.AdaptiveAvgPool2D(output_size=1)
-        self.linear = nn.Linear(int(512 * width_multiplier[3]), class_dim)
-
-    def _make_stage(self, planes, num_blocks, stride):
-        strides = [stride] + [1] * (num_blocks - 1)
-        blocks = []
-        for stride in strides:
-            cur_groups = self.override_groups_map.get(self.cur_layer_idx, 1)
-            blocks.append(
-                RepVGGBlock(
-                    in_channels=self.in_planes,
-                    out_channels=planes,
-                    kernel_size=3,
-                    stride=stride,
-                    padding=1,
-                    groups=cur_groups))
-            self.in_planes = planes
-            self.cur_layer_idx += 1
-        return nn.Sequential(*blocks)
-
-    def eval(self):
-        self.training = False
-        for layer in self.sublayers():
-            layer.training = False
-            layer.eval()
-
-    def forward(self, x):
-        out = self.stage0(x)
-        out = self.stage1(out)
-        out = self.stage2(out)
-        out = self.stage3(out)
-        out = self.stage4(out)
-        out = self.gap(out)
-        out = paddle.flatten(out, start_axis=1)
-        out = self.linear(out)
-        return out
-
-
-optional_groupwise_layers = [2, 4, 6, 8, 10, 12, 14, 16, 18, 20, 22, 24, 26]
-g2_map = {l: 2 for l in optional_groupwise_layers}
-g4_map = {l: 4 for l in optional_groupwise_layers}
-
-
-def RepVGG_A0(**kwargs):
-    return RepVGG(
-        num_blocks=[2, 4, 14, 1],
-        width_multiplier=[0.75, 0.75, 0.75, 2.5],
-        override_groups_map=None,
-        **kwargs)
-
-
-def RepVGG_A1(**kwargs):
-    return RepVGG(
-        num_blocks=[2, 4, 14, 1],
-        width_multiplier=[1, 1, 1, 2.5],
-        override_groups_map=None,
-        **kwargs)
-
-
-def RepVGG_A2(**kwargs):
-    return RepVGG(
-        num_blocks=[2, 4, 14, 1],
-        width_multiplier=[1.5, 1.5, 1.5, 2.75],
-        override_groups_map=None,
-        **kwargs)
-
-
-def RepVGG_B0(**kwargs):
-    return RepVGG(
-        num_blocks=[4, 6, 16, 1],
-        width_multiplier=[1, 1, 1, 2.5],
-        override_groups_map=None,
-        **kwargs)
-
-
-def RepVGG_B1(**kwargs):
-    return RepVGG(
-        num_blocks=[4, 6, 16, 1],
-        width_multiplier=[2, 2, 2, 4],
-        override_groups_map=None,
-        **kwargs)
-
-
-def RepVGG_B1g2(**kwargs):
-    return RepVGG(
-        num_blocks=[4, 6, 16, 1],
-        width_multiplier=[2, 2, 2, 4],
-        override_groups_map=g2_map,
-        **kwargs)
-
-
-def RepVGG_B1g4(**kwargs):
-    return RepVGG(
-        num_blocks=[4, 6, 16, 1],
-        width_multiplier=[2, 2, 2, 4],
-        override_groups_map=g4_map,
-        **kwargs)
-
-
-def RepVGG_B2(**kwargs):
-    return RepVGG(
-        num_blocks=[4, 6, 16, 1],
-        width_multiplier=[2.5, 2.5, 2.5, 5],
-        override_groups_map=None,
-        **kwargs)
-
-
-def RepVGG_B2g2(**kwargs):
-    return RepVGG(
-        num_blocks=[4, 6, 16, 1],
-        width_multiplier=[2.5, 2.5, 2.5, 5],
-        override_groups_map=g2_map,
-        **kwargs)
-
-
-def RepVGG_B2g4(**kwargs):
-    return RepVGG(
-        num_blocks=[4, 6, 16, 1],
-        width_multiplier=[2.5, 2.5, 2.5, 5],
-        override_groups_map=g4_map,
-        **kwargs)
-
-
-def RepVGG_B3(**kwargs):
-    return RepVGG(
-        num_blocks=[4, 6, 16, 1],
-        width_multiplier=[3, 3, 3, 5],
-        override_groups_map=None,
-        **kwargs)
-
-
-def RepVGG_B3g2(**kwargs):
-    return RepVGG(
-        num_blocks=[4, 6, 16, 1],
-        width_multiplier=[3, 3, 3, 5],
-        override_groups_map=g2_map,
-        **kwargs)
-
-
-def RepVGG_B3g4(**kwargs):
-    return RepVGG(
-        num_blocks=[4, 6, 16, 1],
-        width_multiplier=[3, 3, 3, 5],
-        override_groups_map=g4_map,
-        **kwargs)