support repvgg (#429)

Summary:
* support repvgg backbone, and verify the consistency of train mode and eval mode
* onnx export logger style modification

fastreid/modeling/backbones/__init__.py

@@ -13,3 +13,4 @@
 from .regnet import build_regnet_backbone, build_effnet_backbone
 from .shufflenet import build_shufflenetv2_backbone
 from .mobilenet import build_mobilenetv2_backbone
+from .repvgg import build_repvgg_backbone

fastreid/modeling/backbones/regnet/regnet.py

@@ -549,6 +549,10 @@ def build_regnet_backbone(cfg):
     # fmt: on
 
     cfg_files = {
+        '200x': 'fastreid/modeling/backbones/regnet/regnetx/RegNetX-200MF_dds_8gpu.yaml',
+        '200y': 'fastreid/modeling/backbones/regnet/regnety/RegNetY-200MF_dds_8gpu.yaml',
+        '400x': 'fastreid/modeling/backbones/regnet/regnetx/RegNetX-400MF_dds_8gpu.yaml',
+        '400y': 'fastreid/modeling/backbones/regnet/regnety/RegNetY-400MF_dds_8gpu.yaml',
         '800x': 'fastreid/modeling/backbones/regnet/regnetx/RegNetX-800MF_dds_8gpu.yaml',
         '800y': 'fastreid/modeling/backbones/regnet/regnety/RegNetY-800MF_dds_8gpu.yaml',
         '1600x': 'fastreid/modeling/backbones/regnet/regnetx/RegNetX-1.6GF_dds_8gpu.yaml',

fastreid/modeling/backbones/repvgg.py

@@ -0,0 +1,309 @@
+# encoding: utf-8
+# ref: https://github.com/CaoWGG/RepVGG/blob/develop/repvgg.py
+
+
+import logging
+
+import numpy as np
+import torch
+import torch.nn as nn
+
+from fastreid.layers import *
+from fastreid.utils.checkpoint import get_missing_parameters_message, get_unexpected_parameters_message
+from .build import BACKBONE_REGISTRY
+
+logger = logging.getLogger(__name__)
+
+
+def deploy(self, mode=False):
+    self.deploying = mode
+    for module in self.children():
+        if hasattr(module, 'deploying'):
+            module.deploy(mode)
+
+
+nn.Sequential.deploying = False
+nn.Sequential.deploy = deploy
+
+
+def conv_bn(norm_type, in_channels, out_channels, kernel_size, stride, padding, groups=1):
+    result = nn.Sequential()
+    result.add_module('conv', nn.Conv2d(in_channels=in_channels, out_channels=out_channels,
+                                        kernel_size=kernel_size, stride=stride, padding=padding, groups=groups,
+                                        bias=False))
+    result.add_module('bn', get_norm(norm_type, out_channels))
+    return result
+
+
+class RepVGGBlock(nn.Module):
+
+    def __init__(self, in_channels, out_channels, norm_type, kernel_size,
+                 stride=1, padding=0, groups=1):
+        super(RepVGGBlock, self).__init__()
+        self.deploying = False
+
+        self.groups = groups
+        self.in_channels = in_channels
+
+        assert kernel_size == 3
+        assert padding == 1
+
+        padding_11 = padding - kernel_size // 2
+
+        self.nonlinearity = nn.ReLU()
+
+        self.in_channels = in_channels
+        self.in_channels = in_channels
+        self.kernel_size = kernel_size
+        self.stride = stride
+        self.padding = padding
+        self.groups = groups
+
+        self.register_parameter('fused_weight', None)
+        self.register_parameter('fused_bias', None)
+
+        self.rbr_identity = get_norm(norm_type, in_channels) if out_channels == in_channels and stride == 1 else None
+        self.rbr_dense = conv_bn(norm_type, in_channels=in_channels, out_channels=out_channels, kernel_size=kernel_size,
+                                 stride=stride, padding=padding, groups=groups)
+        self.rbr_1x1 = conv_bn(norm_type, in_channels=in_channels, out_channels=out_channels, kernel_size=1,
+                               stride=stride, padding=padding_11, groups=groups)
+
+    def forward(self, inputs):
+        if self.deploying:
+            assert self.fused_weight is not None and self.fused_bias is not None, \
+                "Make deploy mode=True to generate fused weight and fused bias first"
+            fused_out = self.nonlinearity(torch.nn.functional.conv2d(
+                inputs, self.fused_weight, self.fused_bias, self.stride, self.padding, 1, self.groups))
+            return fused_out
+
+        if self.rbr_identity is None:
+            id_out = 0
+        else:
+            id_out = self.rbr_identity(inputs)
+        out = self.nonlinearity(self.rbr_dense(inputs) + self.rbr_1x1(inputs) + id_out)
+
+        return out
+
+    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 torch.nn.functional.pad(kernel1x1, [1, 1, 1, 1])
+
+    def _fuse_bn_tensor(self, branch):
+        if branch is None:
+            return 0, 0
+        if isinstance(branch, nn.Sequential):
+            kernel = branch.conv.weight
+            running_mean = branch.bn.running_mean
+            running_var = branch.bn.running_var
+            gamma = branch.bn.weight
+            beta = branch.bn.bias
+            eps = branch.bn.eps
+        else:
+            assert branch.__class__.__name__.find('BatchNorm') != -1
+            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 = torch.from_numpy(kernel_value).to(branch.weight.device)
+            kernel = self.id_tensor
+            running_mean = branch.running_mean
+            running_var = branch.running_var
+            gamma = branch.weight
+            beta = branch.bias
+            eps = branch.eps
+        std = (running_var + eps).sqrt()
+        t = (gamma / std).reshape(-1, 1, 1, 1)
+        return kernel * t, beta - running_mean * gamma / std
+
+    def deploy(self, mode=False):
+        self.deploying = mode
+        if mode:
+            fused_weight, fused_bias = self.get_equivalent_kernel_bias()
+            self.register_parameter('fused_weight', nn.Parameter(fused_weight))
+            self.register_parameter('fused_bias', nn.Parameter(fused_bias))
+            del self.rbr_identity, self.rbr_1x1, self.rbr_dense
+
+
+class RepVGG(nn.Module):
+
+    def __init__(self, last_stride, norm_type, num_blocks, width_multiplier=None, override_groups_map=None):
+        super(RepVGG, self).__init__()
+
+        assert len(width_multiplier) == 4
+
+        self.deploying = False
+        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, norm_type=norm_type,
+                                  kernel_size=3, stride=2, padding=1)
+        self.cur_layer_idx = 1
+        self.stage1 = self._make_stage(int(64 * width_multiplier[0]), norm_type, num_blocks[0], stride=2)
+        self.stage2 = self._make_stage(int(128 * width_multiplier[1]), norm_type, num_blocks[1], stride=2)
+        self.stage3 = self._make_stage(int(256 * width_multiplier[2]), norm_type, num_blocks[2], stride=2)
+        self.stage4 = self._make_stage(int(512 * width_multiplier[3]), norm_type, num_blocks[3], stride=last_stride)
+
+    def _make_stage(self, planes, norm_type, 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, norm_type=norm_type,
+                                      kernel_size=3, stride=stride, padding=1, groups=cur_groups))
+            self.in_planes = planes
+            self.cur_layer_idx += 1
+        return nn.Sequential(*blocks)
+
+    def deploy(self, mode=False):
+        self.deploying = mode
+        for module in self.children():
+            if hasattr(module, 'deploying'):
+                module.deploy(mode)
+
+    def forward(self, x):
+        out = self.stage0(x)
+        out = self.stage1(out)
+        out = self.stage2(out)
+        out = self.stage3(out)
+        out = self.stage4(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 create_RepVGG_A0(last_stride, norm_type):
+    return RepVGG(last_stride, norm_type, num_blocks=[2, 4, 14, 1],
+                  width_multiplier=[0.75, 0.75, 0.75, 2.5], override_groups_map=None)
+
+
+def create_RepVGG_A1(last_stride, norm_type):
+    return RepVGG(last_stride, norm_type, num_blocks=[2, 4, 14, 1],
+                  width_multiplier=[1, 1, 1, 2.5], override_groups_map=None)
+
+
+def create_RepVGG_A2(last_stride, norm_type):
+    return RepVGG(last_stride, norm_type, num_blocks=[2, 4, 14, 1],
+                  width_multiplier=[1.5, 1.5, 1.5, 2.75], override_groups_map=None)
+
+
+def create_RepVGG_B0(last_stride, norm_type):
+    return RepVGG(last_stride, norm_type, num_blocks=[4, 6, 16, 1],
+                  width_multiplier=[1, 1, 1, 2.5], override_groups_map=None)
+
+
+def create_RepVGG_B1(last_stride, norm_type):
+    return RepVGG(last_stride, norm_type, num_blocks=[4, 6, 16, 1],
+                  width_multiplier=[2, 2, 2, 4], override_groups_map=None)
+
+
+def create_RepVGG_B1g2(last_stride, norm_type):
+    return RepVGG(last_stride, norm_type, num_blocks=[4, 6, 16, 1],
+                  width_multiplier=[2, 2, 2, 4], override_groups_map=g2_map)
+
+
+def create_RepVGG_B1g4(last_stride, norm_type):
+    return RepVGG(last_stride, norm_type, num_blocks=[4, 6, 16, 1],
+                  width_multiplier=[2, 2, 2, 4], override_groups_map=g4_map)
+
+
+def create_RepVGG_B2(last_stride, norm_type):
+    return RepVGG(last_stride, norm_type, num_blocks=[4, 6, 16, 1],
+                  width_multiplier=[2.5, 2.5, 2.5, 5], override_groups_map=None)
+
+
+def create_RepVGG_B2g2(last_stride, norm_type):
+    return RepVGG(last_stride, norm_type, num_blocks=[4, 6, 16, 1],
+                  width_multiplier=[2.5, 2.5, 2.5, 5], override_groups_map=g2_map)
+
+
+def create_RepVGG_B2g4(last_stride, norm_type):
+    return RepVGG(last_stride, norm_type, num_blocks=[4, 6, 16, 1],
+                  width_multiplier=[2.5, 2.5, 2.5, 5], override_groups_map=g4_map)
+
+
+def create_RepVGG_B3(last_stride, norm_type):
+    return RepVGG(last_stride, norm_type, num_blocks=[4, 6, 16, 1],
+                  width_multiplier=[3, 3, 3, 5], override_groups_map=None)
+
+
+def create_RepVGG_B3g2(last_stride, norm_type):
+    return RepVGG(last_stride, norm_type, num_blocks=[4, 6, 16, 1],
+                  width_multiplier=[3, 3, 3, 5], override_groups_map=g2_map)
+
+
+def create_RepVGG_B3g4(last_stride, norm_type):
+    return RepVGG(last_stride, norm_type, num_blocks=[4, 6, 16, 1],
+                  width_multiplier=[3, 3, 3, 5], override_groups_map=g4_map)
+
+
+@BACKBONE_REGISTRY.register()
+def build_repvgg_backbone(cfg):
+    """
+    Create a RepVGG instance from config.
+    Returns:
+        RepVGG: a :class: `RepVGG` instance.
+    """
+
+    # fmt: off
+    pretrain      = cfg.MODEL.BACKBONE.PRETRAIN
+    pretrain_path = cfg.MODEL.BACKBONE.PRETRAIN_PATH
+    last_stride   = cfg.MODEL.BACKBONE.LAST_STRIDE
+    bn_norm       = cfg.MODEL.BACKBONE.NORM
+    depth         = cfg.MODEL.BACKBONE.DEPTH
+    # fmt: on
+
+    func_dict = {
+        'A0': create_RepVGG_A0,
+        'A1': create_RepVGG_A1,
+        'A2': create_RepVGG_A2,
+        'B0': create_RepVGG_B0,
+        'B1': create_RepVGG_B1,
+        'B1g2': create_RepVGG_B1g2,
+        'B1g4': create_RepVGG_B1g4,
+        'B2': create_RepVGG_B2,
+        'B2g2': create_RepVGG_B2g2,
+        'B2g4': create_RepVGG_B2g4,
+        'B3': create_RepVGG_B3,
+        'B3g2': create_RepVGG_B3g2,
+        'B3g4': create_RepVGG_B3g4,
+    }
+
+    model = func_dict[depth](last_stride, bn_norm)
+
+    if pretrain:
+        try:
+            state_dict = torch.load(pretrain_path, map_location=torch.device("cpu"))
+            logger.info(f"Loading pretrained model from {pretrain_path}")
+        except FileNotFoundError as e:
+            logger.info(f'{pretrain_path} is not found! Please check this path.')
+            raise e
+        except KeyError as e:
+            logger.info("State dict keys error! Please check the state dict.")
+            raise e
+
+        incompatible = model.load_state_dict(state_dict, strict=False)
+        if incompatible.missing_keys:
+            logger.info(
+                get_missing_parameters_message(incompatible.missing_keys)
+            )
+        if incompatible.unexpected_keys:
+            logger.info(
+                get_unexpected_parameters_message(incompatible.unexpected_keys)
+            )
+
+    return model

tests/test_repvgg.py

@@ -0,0 +1,33 @@
+import sys
+import unittest
+
+import torch
+
+sys.path.append('.')
+from fastreid.config import get_cfg
+from fastreid.modeling.backbones import build_backbone
+
+
+class MyTestCase(unittest.TestCase):
+    def test_fusebn(self):
+        cfg = get_cfg()
+        cfg.defrost()
+        cfg.MODEL.BACKBONE.NAME = 'build_repvgg_backbone'
+        cfg.MODEL.BACKBONE.DEPTH = 'B1g2'
+        cfg.MODEL.BACKBONE.PRETRAIN = False
+        model = build_backbone(cfg)
+        model.eval()
+
+        test_inp = torch.randn((1, 3, 256, 128))
+
+        y = model(test_inp)
+
+        model.deploy(mode=True)
+        from ipdb import set_trace; set_trace()
+        fused_y = model(test_inp)
+
+        print("final error :", torch.max(torch.abs(fused_y - y)).item())
+
+
+if __name__ == '__main__':
+    unittest.main()