support resnet50

examples/oneflow2onnx/models/test_resnet50.py

@@ -13,182 +13,305 @@ 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 oneflow as flow
-import oneflow.typing as tp
-import onnx
-import onnxruntime as ort
-import numpy as np
+import oneflow.nn as nn
+from oneflow import Tensor
+from typing import Type, Any, Callable, Union, List, Optional
 from oneflow_onnx.oneflow2onnx.util import convert_to_onnx_and_check
-
-
-BLOCK_COUNTS = [3, 4, 6, 3]
-BLOCK_FILTERS = [256, 512, 1024, 2048]
-BLOCK_FILTERS_INNER = [64, 128, 256, 512]
-
-g_trainable = False
-
-
-def _conv2d(
-    name,
-    input,
-    filters,
-    kernel_size,
-    strides=1,
-    padding="SAME",
-    data_format="NCHW",
-    dilations=1,
-    trainable=True,
-    # weight_initializer=flow.variance_scaling_initializer(data_format="NCHW"),
-    weight_initializer=flow.variance_scaling_initializer(
-        2, "fan_in", "random_normal", data_format="NCHW"
-    ),
-    weight_regularizer=flow.regularizers.l2(1.0 / 32768),
-):
-    weight = flow.get_variable(
-        name + "-weight",
-        shape=(filters, input.shape[1], kernel_size, kernel_size),
-        dtype=input.dtype,
-        initializer=weight_initializer,
-        regularizer=weight_regularizer,
-        model_name="weight",
-        trainable=trainable,
-    )
-    return flow.nn.conv2d(
-        input, weight, strides, padding, data_format, dilations, name=name
-    )
-
-
-def _batch_norm(inputs, name=None, trainable=True):
-    return flow.layers.batch_normalization(
-        inputs=inputs,
-        axis=1,
-        momentum=0.9,  # 97,
-        epsilon=1.001e-5,
-        center=True,
-        scale=True,
-        trainable=trainable,
-        training=trainable,
-        name=name,
-    )
-
-
-def conv2d_affine(input, name, filters, kernel_size, strides, activation=None):
-    # input data_format must be NCHW, cannot check now
-    padding = "SAME" if strides > 1 or kernel_size > 1 else "VALID"
-    output = _conv2d(
-        name, input, filters, kernel_size, strides, padding, trainable=g_trainable
-    )
-    output = _batch_norm(output, name + "_bn", trainable=g_trainable)
-    if activation == "Relu":
-        output = flow.math.relu(output)
-
-    return output
-
-
-def bottleneck_transformation(input, block_name, filters, filters_inner, strides):
-    a = conv2d_affine(
-        input, block_name + "_branch2a", filters_inner, 1, 1, activation="Relu",
-    )
-
-    b = conv2d_affine(
-        a, block_name + "_branch2b", filters_inner, 3, strides, activation="Relu",
+import tempfile
+def conv3x3(
+    in_planes: int, out_planes: int, stride: int = 1, groups: int = 1, dilation: int = 1
+) -> nn.Conv2d:
+    """3x3 convolution with padding"""
+    return nn.Conv2d(
+        in_planes,
+        out_planes,
+        kernel_size=3,
+        stride=stride,
+        padding=dilation,
+        groups=groups,
+        bias=False,
+        dilation=dilation,
     )
 
-    c = conv2d_affine(b, block_name + "_branch2c", filters, 1, 1)
 
-    return c
-
-
-def residual_block(input, block_name, filters, filters_inner, strides_init):
-    if strides_init != 1 or block_name == "res2_0":
-        shortcut = conv2d_affine(
-            input, block_name + "_branch1", filters, 1, strides_init
+def conv1x1(in_planes: int, out_planes: int, stride: int = 1) -> nn.Conv2d:
+    """1x1 convolution"""
+    return nn.Conv2d(in_planes, out_planes, kernel_size=1, stride=stride, bias=False)
+
+
+class BasicBlock(nn.Module):
+    expansion: int = 1
+
+    def __init__(
+        self,
+        inplanes: int,
+        planes: int,
+        stride: int = 1,
+        downsample: Optional[nn.Module] = None,
+        groups: int = 1,
+        base_width: int = 64,
+        dilation: int = 1,
+        norm_layer: Optional[Callable[..., nn.Module]] = None,
+    ) -> None:
+        super(BasicBlock, self).__init__()
+        if norm_layer is None:
+            norm_layer = nn.BatchNorm2d
+        if groups != 1 or base_width != 64:
+            raise ValueError("BasicBlock only supports groups=1 and base_width=64")
+        if dilation > 1:
+            raise NotImplementedError("Dilation > 1 not supported in BasicBlock")
+        # Both self.conv1 and self.downsample layers downsample the input when stride != 1
+        self.conv1 = conv3x3(inplanes, planes, stride)
+        self.bn1 = norm_layer(planes)
+        self.relu = nn.ReLU()
+        self.conv2 = conv3x3(planes, planes)
+        self.bn2 = norm_layer(planes)
+        self.downsample = downsample
+        self.stride = stride
+
+    def forward(self, x: Tensor) -> Tensor:
+        identity = x
+
+        out = self.conv1(x)
+        out = self.bn1(out)
+        out = self.relu(out)
+
+        out = self.conv2(out)
+        out = self.bn2(out)
+
+        if self.downsample is not None:
+            identity = self.downsample(x)
+
+        out += identity
+        out = self.relu(out)
+
+        return out
+
+
+class Bottleneck(nn.Module):
+    expansion: int = 4
+
+    def __init__(
+        self,
+        inplanes: int,
+        planes: int,
+        stride: int = 1,
+        downsample: Optional[nn.Module] = None,
+        groups: int = 1,
+        base_width: int = 64,
+        dilation: int = 1,
+        norm_layer: Optional[Callable[..., nn.Module]] = None,
+    ) -> None:
+        super(Bottleneck, self).__init__()
+        if norm_layer is None:
+            norm_layer = nn.BatchNorm2d
+        width = int(planes * (base_width / 64.0)) * groups
+        # Both self.conv2 and self.downsample layers downsample the input when stride != 1
+        self.conv1 = conv1x1(inplanes, width)
+        self.bn1 = norm_layer(width)
+        self.conv2 = conv3x3(width, width, stride, groups, dilation)
+        self.bn2 = norm_layer(width)
+        self.conv3 = conv1x1(width, planes * self.expansion)
+        self.bn3 = norm_layer(planes * self.expansion)
+        self.relu = nn.ReLU()
+        self.downsample = downsample
+        self.stride = stride
+
+    def forward(self, x: Tensor) -> Tensor:
+        identity = x
+
+        out = self.conv1(x)
+        out = self.bn1(out)
+        out = self.relu(out)
+
+        out = self.conv2(out)
+        out = self.bn2(out)
+        out = self.relu(out)
+
+        out = self.conv3(out)
+        out = self.bn3(out)
+
+        if self.downsample is not None:
+            identity = self.downsample(x)
+
+        out += identity
+        out = self.relu(out)
+
+        return out
+
+
+class ResNet(nn.Module):
+    def __init__(
+        self,
+        block: Type[Union[BasicBlock, Bottleneck]],
+        layers: List[int],
+        num_classes: int = 1000,
+        zero_init_residual: bool = False,
+        groups: int = 1,
+        width_per_group: int = 64,
+        replace_stride_with_dilation: Optional[List[bool]] = None,
+        norm_layer: Optional[Callable[..., nn.Module]] = None,
+    ) -> None:
+        super(ResNet, self).__init__()
+        if norm_layer is None:
+            norm_layer = nn.BatchNorm2d
+        self._norm_layer = norm_layer
+
+        self.inplanes = 64
+        self.dilation = 1
+        if replace_stride_with_dilation is None:
+            # each element in the tuple indicates if we should replace
+            # the 2x2 stride with a dilated convolution instead
+            replace_stride_with_dilation = [False, False, False]
+        if len(replace_stride_with_dilation) != 3:
+            raise ValueError(
+                "replace_stride_with_dilation should be None "
+                "or a 3-element tuple, got {}".format(replace_stride_with_dilation)
+            )
+        self.groups = groups
+        self.base_width = width_per_group
+        self.conv1 = nn.Conv2d(
+            3, self.inplanes, kernel_size=7, stride=2, padding=3, bias=False
         )
-    else:
-        shortcut = input
-
-    bottleneck = bottleneck_transformation(
-        input, block_name, filters, filters_inner, strides_init
-    )
-
-    return flow.math.relu(bottleneck + shortcut)
-
-
-def residual_stage(input, stage_name, counts, filters, filters_inner, stride_init=2):
-    output = input
-    for i in range(counts):
-        block_name = "%s_%d" % (stage_name, i)
-        output = residual_block(
-            output, block_name, filters, filters_inner, stride_init if i == 0 else 1,
+        self.bn1 = norm_layer(self.inplanes)
+        self.relu = nn.ReLU()
+        self.maxpool = nn.MaxPool2d(kernel_size=3, stride=2, padding=1)
+        self.layer1 = self._make_layer(block, 64, layers[0])
+        self.layer2 = self._make_layer(
+            block, 128, layers[1], stride=2, dilate=replace_stride_with_dilation[0]
         )
-
-    return output
-
-
-def resnet_conv_x_body(input, on_stage_end=lambda x: x):
-    output = input
-    for i, (counts, filters, filters_inner) in enumerate(
-        zip(BLOCK_COUNTS, BLOCK_FILTERS, BLOCK_FILTERS_INNER)
-    ):
-        stage_name = "res%d" % (i + 2)
-        output = residual_stage(
-            output, stage_name, counts, filters, filters_inner, 1 if i == 0 else 2,
+        self.layer3 = self._make_layer(
+            block, 256, layers[2], stride=2, dilate=replace_stride_with_dilation[1]
         )
-        on_stage_end(output)
-
-    return output
-
-
-def resnet_stem(input):
-    conv1 = _conv2d("conv1", input, 1, 1, 2)
-    tmp = _batch_norm(conv1, "conv1_bn", trainable=g_trainable)
-    conv1_bn = flow.math.relu(tmp)
-    pool1 = flow.nn.max_pool2d(
-        conv1_bn, ksize=3, strides=2, padding="VALID", data_format="NCHW", name="pool1",
-    )
-    return pool1
-
-
-def resnet50(images, trainable=True, need_transpose=False):
-
-    # note: images.shape = (N C H W) in cc's new dataloader, transpose is not needed anymore
-    if need_transpose:
-        images = flow.transpose(images, name="transpose", perm=[0, 3, 1, 2])
-
-    with flow.scope.namespace("Resnet"):
-        stem = resnet_stem(images)
-        body = resnet_conv_x_body(stem, lambda x: x)
-        pool5 = flow.nn.avg_pool2d(
-            body, ksize=7, strides=1, padding="VALID", data_format="NCHW", name="pool5",
+        self.layer4 = self._make_layer(
+            block, 512, layers[3], stride=2, dilate=replace_stride_with_dilation[2]
         )
-
-        fc1001 = flow.layers.dense(
-            flow.reshape(pool5, (pool5.shape[0], -1)),
-            units=1000,
-            use_bias=True,
-            kernel_initializer=flow.variance_scaling_initializer(
-                2, "fan_in", "random_normal"
-            ),
-            # kernel_initializer=flow.xavier_uniform_initializer(),
-            bias_initializer=flow.random_uniform_initializer(),
-            kernel_regularizer=flow.regularizers.l2(1.0 / 32768),
-            trainable=trainable,
-            name="fc1001",
+        self.avgpool = nn.AvgPool2d((7, 7))
+        self.fc = nn.Linear(512 * block.expansion, num_classes)
+
+        for m in self.modules():
+            if isinstance(m, nn.Conv2d):
+                nn.init.kaiming_normal_(m.weight, mode="fan_out", nonlinearity="relu")
+            elif isinstance(m, nn.BatchNorm2d):
+                nn.init.constant_(m.weight, 1)
+                nn.init.constant_(m.bias, 0)
+
+        # Zero-initialize the last BN in each residual branch,
+        # so that the residual branch starts with zeros, and each residual block behaves like an identity.
+        # This improves the model by 0.2~0.3% according to https://arxiv.org/abs/1706.02677
+        if zero_init_residual:
+            for m in self.modules():
+                if isinstance(m, Bottleneck):
+                    nn.init.constant_(m.bn3.weight, 0)  # type: ignore[arg-type]
+                elif isinstance(m, BasicBlock):
+                    nn.init.constant_(m.bn2.weight, 0)  # type: ignore[arg-type]
+
+    def _make_layer(
+        self,
+        block: Type[Union[BasicBlock, Bottleneck]],
+        planes: int,
+        blocks: int,
+        stride: int = 1,
+        dilate: bool = False,
+    ) -> nn.Sequential:
+        norm_layer = self._norm_layer
+        downsample = None
+        previous_dilation = self.dilation
+        if dilate:
+            self.dilation *= stride
+            stride = 1
+        if stride != 1 or self.inplanes != planes * block.expansion:
+            downsample = nn.Sequential(
+                conv1x1(self.inplanes, planes * block.expansion, stride),
+                norm_layer(planes * block.expansion),
+            )
+
+        layers = []
+        layers.append(
+            block(
+                self.inplanes,
+                planes,
+                stride,
+                downsample,
+                self.groups,
+                self.base_width,
+                previous_dilation,
+                norm_layer,
+            )
         )
-
-    return fc1001
-
-
-def test_resnet50():
-    @flow.global_function()
-    def InferenceNet(images: tp.Numpy.Placeholder((1, 3, 224, 224))):
-        logits = resnet50(images)
-
-        predictions = flow.nn.softmax(logits)
-        return predictions
-
-    convert_to_onnx_and_check(InferenceNet, flow_weight_dir=None, onnx_model_path="/tmp")
+        self.inplanes = planes * block.expansion
+        for _ in range(1, blocks):
+            layers.append(
+                block(
+                    self.inplanes,
+                    planes,
+                    groups=self.groups,
+                    base_width=self.base_width,
+                    dilation=self.dilation,
+                    norm_layer=norm_layer,
+                )
+            )
+
+        return nn.Sequential(*layers)
+
+    def _forward_impl(self, x: Tensor) -> Tensor:
+        x = self.conv1(x)
+        x = self.bn1(x)
+        x = self.relu(x)
+        x = self.maxpool(x)
+
+        x = self.layer1(x)
+        x = self.layer2(x)
+        x = self.layer3(x)
+        x = self.layer4(x)
+
+        x = self.avgpool(x)
+        x = flow.flatten(x, 1)
+        x = self.fc(x)
+
+        return x
+
+    def forward(self, x: Tensor) -> Tensor:
+        return self._forward_impl(x)
+
+
+def _resnet(
+    arch: str,
+    block: Type[Union[BasicBlock, Bottleneck]],
+    layers: List[int],
+    **kwargs: Any
+) -> ResNet:
+    model = ResNet(block, layers, **kwargs)
+    return model
+
+
+def resnet50(**kwargs: Any) -> ResNet:
+    r"""ResNet-5
+    `"Deep Residual Learning for Image Recognition" <https://arxiv.org/pdf/1512.03385.pdf>`_.
+    """
+    return _resnet("resnet50", Bottleneck, [3, 4, 6, 3], **kwargs)
+
+resnet = resnet50()
+resnet = resnet.to("cuda")
+resnet.eval()
+class ResNetGraph(flow.nn.Graph):
+    def __init__(self):
+        super().__init__()
+        self.m = resnet
+
+    def build(self, x):
+        out = self.m(x)
+        return out
+
+def test_resnet():
+    
+    resnet_graph = ResNetGraph()
+    resnet_graph._compile(flow.randn(1, 3, 224, 224).to("cuda"))
+
+    with tempfile.TemporaryDirectory() as tmpdirname:
+        flow.save(resnet.state_dict(), tmpdirname)
+        convert_to_onnx_and_check(resnet_graph, flow_weight_dir=tmpdirname, onnx_model_path="/tmp", print_outlier=False)
+
+test_resnet()

oneflow_onnx/oneflow2onnx/flow2onnx.py

@@ -86,7 +86,9 @@ def FlowToOnnxNaive(graph, shape_override):
                 for order in node.user_conf.input_order:
                     for key, val in node.user_conf.input.items():
                         if key == order:
-                            res.append(val.s[0])
+                            for _ in range(len(val.s)):
+                                res.append(val.s[_])
+
                 return res
             ipts = []
             for ibn in ibns:
@@ -122,7 +124,8 @@ def FlowToOnnxNaive(graph, shape_override):
                 for order in node.user_conf.output_order:
                     for key, val in node.user_conf.output.items():
                         if key == order:
-                            res.append(val.s[0])
+                            for _ in range(len(val.s)):
+                                res.append(val.s[_])
                 return res
             outputs = []
             for obn in obns:
@@ -169,8 +172,6 @@ def FlowToOnnxNaive(graph, shape_override):
             op_type = get_op_type(node)
             input_names = get_inputs(node)
             output_names = get_outputs(node)
-            input_order = node.user_conf.input_order
-            output_order = node.user_conf.output_order
             onnx_node = helper.make_node(
                 op_type, input_names, output_names, name=node.name, **attr 
             )