以img_backbone(Resnet50为例)
1 先拿到pytorch版的参数
./tools/dist_train.sh ./projects/configs/bevformer/bevformer_tiny.py
在bevformer.py文件__init__之后打断点,通过self.img_backbone.state_dict()得到入参数--
是一个<class 'collections.OrderedDict'>这种格式的数据;
self.img_backbone.state_dict().keys(),可以得到全部的参数
写入一个文件里面得到pytorchKey.txt
2 拿到mindspore版的参数
在resnet.py这个文件里面
import warnings
from mindspore import nn
class ResLayer(nn.SequentialCell):
"""ResLayer to build ResNet style backbone.
Args:
block (nn.Module): block used to build ResLayer.
inplanes (int): inplanes of block.
planes (int): planes of block.
num_blocks (int): number of blocks.
stride (int): stride of the first block. Default: 1
avg_down (bool): Use AvgPool instead of stride conv when
downsampling in the bottleneck. Default: False
conv_cfg (dict): dictionary to construct and config conv layer.
Default: None
norm_cfg (dict): dictionary to construct and config norm layer.
Default: dict(type='BN')
downsample_first (bool): Downsample at the first block or last block.
False for Hourglass, True for ResNet. Default: True
"""
def __init__(self,
block,
inplanes,
planes,
num_blocks,
stride=1,
avg_down=False,
conv_cfg=None,
norm_cfg=dict(type='BN'),
downsample_first=True,
**kwargs):
self.block = block
downsample = None
if stride != 1 or inplanes != planes * block.expansion:
downsample = []
conv_stride = stride
if avg_down:
conv_stride = 1
downsample.append(
nn.AvgPool2d(
kernel_size=stride,
stride=stride,
ceil_mode=True,
count_include_pad=False))
downsample.extend([
nn.Conv2d(
in_channels=inplanes,
out_channels=planes * block.expansion,
kernel_size=1,
stride=conv_stride,
has_bias=False),
nn.BatchNorm2d(num_features=planes * block.expansion)
])
downsample = nn.SequentialCell(*downsample)
layers = []
if downsample_first:
layers.append(
block(
inplanes=inplanes,
planes=planes,
stride=stride,
downsample=downsample,
conv_cfg=conv_cfg,
norm_cfg=norm_cfg,
**kwargs))
inplanes = planes * block.expansion
for _ in range(1, num_blocks):
layers.append(
block(
inplanes=inplanes,
planes=planes,
stride=1,
conv_cfg=conv_cfg,
norm_cfg=norm_cfg,
**kwargs))
else: # downsample_first=False is for HourglassModule
for _ in range(num_blocks - 1):
layers.append(
block(
inplanes=inplanes,
planes=inplanes,
stride=1,
conv_cfg=conv_cfg,
norm_cfg=norm_cfg,
**kwargs))
layers.append(
block(
inplanes=inplanes,
planes=planes,
stride=stride,
downsample=downsample,
conv_cfg=conv_cfg,
norm_cfg=norm_cfg,
**kwargs))
super(ResLayer, self).__init__(*layers)
class Bottleneck(nn.Cell):
expansion = 4
def __init__(self,
inplanes,
planes,
stride=1,
dilation=1,
downsample=None,
style='pytorch',
with_cp=False,
conv_cfg=None,
norm_cfg=dict(type='BN'),
dcn=None,
plugins=None,
init_cfg=None):
"""Bottleneck block for ResNet.
If style is "pytorch", the stride-two layer is the 3x3 conv layer, if
it is "caffe", the stride-two layer is the first 1x1 conv layer.
"""
super(Bottleneck, self).__init__()
assert style in ['pytorch', 'caffe']
assert dcn is None or isinstance(dcn, dict)
assert plugins is None or isinstance(plugins, list)
if plugins is not None:
allowed_position = ['after_conv1', 'after_conv2', 'after_conv3']
assert all(p['position'] in allowed_position for p in plugins)
self.inplanes = inplanes
self.planes = planes
self.stride = stride
self.dilation = dilation
self.style = style
self.with_cp = with_cp
self.conv_cfg = conv_cfg
self.norm_cfg = norm_cfg
self.dcn = dcn
self.with_dcn = dcn is not None
self.plugins = plugins
self.with_plugins = plugins is not None
if self.with_plugins:
# collect plugins for conv1/conv2/conv3
pass
if self.style == 'pytorch':
self.conv1_stride = 1
self.conv2_stride = stride
else:
self.conv1_stride = stride
self.conv2_stride = 1
if norm_cfg['type'] == 'BN':
self.norm1 = nn.BatchNorm2d(num_features=planes)
self.norm2 = nn.BatchNorm2d(num_features=planes)
self.norm3 = nn.BatchNorm2d(num_features=planes * self.expansion)
self.conv1 = nn.Conv2d(
in_channels=inplanes,
out_channels=planes,
kernel_size=1,
stride=self.conv1_stride,
has_bias=False)
fallback_on_stride = False
if self.with_dcn:
pass
if not self.with_dcn or fallback_on_stride:
self.conv2 = nn.Conv2d(
in_channels=planes,
out_channels=planes,
kernel_size=3,
stride=self.conv2_stride,
padding=dilation,
dilation=dilation,
has_bias=False,
pad_mode='pad')
else:
pass
self.conv3 = nn.Conv2d(
in_channels=planes,
out_channels=planes * self.expansion,
kernel_size=1,
has_bias=False)
self.relu = nn.ReLU()
self.downsample = downsample
def construct(self, x):
"""Forward function."""
identity = x
out = self.conv1(x)
out = self.norm1(out)
out = self.relu(out)
out = self.conv2(out)
out = self.norm2(out)
out = self.relu(out)
out = self.conv3(out)
out = self.norm3(out)
if self.downsample is not None:
identity = self.downsample(x)
out += identity
out = self.relu(out)
return out
class ResNet(nn.Cell):
"""ResNet backbone.
Args:
depth (int): Depth of resnet, from {18, 34, 50, 101, 152}.
stem_channels (int | None): Number of stem channels. If not specified,
it will be the same as `base_channels`. Default: None.
base_channels (int): Number of base channels of res layer. Default: 64.
in_channels (int): Number of input image channels. Default: 3.
num_stages (int): Resnet stages. Default: 4.
strides (Sequence[int]): Strides of the first block of each stage.
dilations (Sequence[int]): Dilation of each stage.
out_indices (Sequence[int]): Output from which stages.
style (str): `pytorch` or `caffe`. If set to "pytorch", the stride-two
layer is the 3x3 conv layer, otherwise the stride-two layer is
the first 1x1 conv layer.
deep_stem (bool): Replace 7x7 conv in input stem with 3 3x3 conv
avg_down (bool): Use AvgPool instead of stride conv when
downsampling in the bottleneck.
frozen_stages (int): Stages to be frozen (stop grad and set eval mode).
-1 means not freezing any parameters.
norm_cfg (dict): Dictionary to construct and config norm layer.
norm_eval (bool): Whether to set norm layers to eval mode, namely,
freeze running stats (mean and var). Note: Effect on Batch Norm
and its variants only.
plugins (list[dict]): List of plugins for stages, each dict contains:
- cfg (dict, required): Cfg dict to build plugin.
- position (str, required): Position inside block to insert
plugin, options are 'after_conv1', 'after_conv2', 'after_conv3'.
- stages (tuple[bool], optional): Stages to apply plugin, length
should be same as 'num_stages'.
with_cp (bool): Use checkpoint or not. Using checkpoint will save some
memory while slowing down the training speed.
zero_init_residual (bool): Whether to use zero init for last norm layer
in resblocks to let them behave as identity.
pretrained (str, optional): model pretrained path. Default: None
init_cfg (dict or list[dict], optional): Initialization config dict.
Default: None
"""
arch_settings = {
50: (Bottleneck, (3, 4, 6, 3)),
101: (Bottleneck, (3, 4, 23, 3)),
152: (Bottleneck, (3, 8, 36, 3))
}
def __init__(self,
depth,
in_channels=3,
stem_channels=None,
base_channels=64,
num_stages=4,
strides=(1, 2, 2, 2),
dilations=(1, 1, 1, 1),
out_indices=(0, 1, 2, 3),
style='pytorch',
deep_stem=False,
avg_down=False,
frozen_stages=-1,
conv_cfg=None,
norm_cfg=dict(type='BN'),
norm_eval=True,
dcn=None,
stage_with_dcn=(False, False, False, False),
plugins=None,
with_cp=False,
zero_init_residual=True,
pretrained=None,
init_cfg=None):
super(ResNet, self).__init__()
self.zero_init_residual = zero_init_residual
if depth not in self.arch_settings:
raise KeyError(f'invalid depth {depth} for resnet')
block_init_cfg = None
assert not (init_cfg and pretrained), \
'init_cfg and pretrained cannot be setting at the same time'
if isinstance(pretrained, str):
warnings.warn('DeprecationWarning: pretrained is deprecated, '
'please use "init_cfg" instead')
self.init_cfg = dict(type='Pretrained', checkpoint=pretrained)
elif pretrained is None:
if init_cfg is None:
self.init_cfg = [
dict(type='Kaiming', layer='Conv2d'),
dict(
type='Constant',
val=1,
layer=['_BatchNorm', 'GroupNorm'])
]
block = self.arch_settings[depth][0]
if self.zero_init_residual:
if block is Bottleneck:
block_init_cfg = dict(
type='Constant',
val=0,
override=dict(name='norm3'))
else:
raise TypeError('pretrained must be a str or None')
self.depth = depth
if stem_channels is None:
stem_channels = base_channels
self.stem_channels = stem_channels
self.base_channels = base_channels
self.num_stages = num_stages
assert num_stages >= 1 and num_stages <= 4
self.strides = strides
self.dilations = dilations
assert len(strides) == len(dilations) == num_stages
self.out_indices = out_indices
assert max(out_indices) < num_stages
self.style = style
self.deep_stem = deep_stem
self.avg_down = avg_down
self.frozen_stages = frozen_stages
self.conv_cfg = conv_cfg
self.norm_cfg = norm_cfg
self.with_cp = with_cp
self.norm_eval = norm_eval
self.dcn = dcn
self.stage_with_dcn = stage_with_dcn
if dcn is not None:
assert len(stage_with_dcn) == num_stages
self.plugins = plugins
self.block, stage_blocks = self.arch_settings[depth]
self.stage_blocks = stage_blocks[:num_stages]
self.inplanes = stem_channels
self._make_stem_layer(in_channels, stem_channels)
self.res_layers = []
for i, num_blocks in enumerate(self.stage_blocks):
stride = strides[i]
dilation = dilations[i]
dcn = self.dcn if self.stage_with_dcn[i] else None
if plugins is None:
stage_plugins = None
planes = base_channels * 2 ** i
res_layer = self.make_res_layer(
block=self.block,
inplanes=self.inplanes,
planes=planes,
num_blocks=num_blocks,
stride=stride,
dilation=dilation,
style=self.style,
avg_down=self.avg_down,
with_cp=with_cp,
conv_cfg=conv_cfg,
norm_cfg=norm_cfg,
dcn=dcn,
plugins=stage_plugins,
init_cfg=block_init_cfg)
self.inplanes = planes * self.block.expansion
layer_name = f'layer{i + 1}'
setattr(self, layer_name, res_layer)
self.res_layers.append(layer_name)
self._freeze_stages() #TODO
self.feat_dim = self.block.expansion * base_channels * 2 ** (
len(self.stage_blocks) - 1)
def make_res_layer(self, **kwargs):
"""Pack all blocks in a stage into a ``ResLayer``."""
return ResLayer(**kwargs)
def _make_stem_layer(self, in_channels, stem_channels):
if self.deep_stem:
self.stem = nn.SequentialCell(
nn.Conv2d(
in_channels=in_channels,
out_channels=stem_channels // 2,
kernel_size=3,
stride=2,
padding=1,
has_bias=False,
pad_mode='pad'),
nn.BatchNorm2d(num_features=stem_channels // 2),
nn.ReLU(),
nn.Conv2d(
in_channels=stem_channels // 2,
out_channels=stem_channels // 2,
kernel_size=3,
stride=1,
padding=1,
has_bias=False,
pad_mode='pad'),
nn.BatchNorm2d(num_features=stem_channels // 2),
nn.ReLU(),
nn.Conv2d(
in_channels=stem_channels // 2,
out_channels=stem_channels,
kernel_size=3,
stride=1,
padding=1,
has_bias=False,
pad_mode='pad'),
nn.BatchNorm2d(num_features=stem_channels),
nn.ReLU())
else:
self.conv1 = nn.Conv2d(
in_channels=in_channels,
out_channels=stem_channels,
kernel_size=7,
stride=2,
padding=3,
has_bias=False,
pad_mode='pad')
self.norm1 = nn.BatchNorm2d(stem_channels)
self.relu = nn.ReLU()
self.maxpool = nn.MaxPool2d(kernel_size=3, stride=2, padding=1, pad_mode='pad')
def _freeze_stages(self):
if self.frozen_stages >= 0:
if self.deep_stem:
self.stem.set_train(False)
for param in self.stem.trainable_params():
param.requires_grad = False
else:
self.norm1.set_train(False)
for m in [self.conv1, self.norm1]:
for param in m.trainable_params():
param.requires_grad = False
for i in range(1, self.frozen_stages + 1):
m = getattr(self, f'layer{i}')
m.set_train(False)
for param in m.trainable_params():
param.requires_grad = False
def construct(self, x, mode=True):
"""Forward function."""
self.train(mode)
if self.deep_stem:
x = self.stem(x)
else:
x = self.conv1(x)
x = self.norm1(x)
x = self.relu(x)
x = self.maxpool(x)
outs = []
for i, layer_name in enumerate(self.res_layers):
res_layer = getattr(self, layer_name)
x = res_layer(x)
if i in self.out_indices:
outs.append(x)
return tuple(outs)
def train(self, mode=True):
"""Convert the model into training mode while keep normalization layer
freezed."""
super(ResNet, self).set_train(mode)
self._freeze_stages()
if mode and self.norm_eval:
for name, module in self.name_cells().items():
# trick: eval have effect on BatchNorm only
if isinstance(module, nn.BatchNorm2d):
module.set_train(False)
param_list = []
net = ResNet(depth=50,
num_stages=4,
out_indices=(3,),
frozen_stages=0,
norm_cfg=dict(type='BN', requires_grad=False),
norm_eval=True,
style='pytorch')
for x in net.get_parameters():
param_list.append(x.name)
print(param_list)
print(len(param_list))
for i in range(len(param_list)):
with open("./mindsporeKey.txt", 'a') as file:
file.write(param_list[i])
file.write("\n")
直接运行一下,重点是net.get_parameters()这个方法可以直接拿到参数
然后写入mindposreKey.txt
3 用beyond compare对比这两个文件,然后根据规律写一个脚本
import json
# 读取第一个文本文件并将内容存储为列表
with open('pytorchKeyBEV.txt', 'r') as file2:
file1_lines = file2.read().splitlines()
# 读取第二个文本文件并将内容存储为列表
with open('mindsporeKey.txt', 'r') as file1:
file2_lines = file1.read().splitlines()
# 初始化一个字典来存储映射关系
mapping = {}
for i in range(len(file1_lines)):
temp = file1_lines[i]
if ".downsample.0" in file1_lines[i] or ("conv" in file1_lines[i]):
mapping[file1_lines[i]] = temp
continue
if "bn" in file1_lines[i]:
temp = temp.replace("bn", "norm")
if "weight" in file1_lines[i]:
temp = temp.replace("weight", "gamma")
if "bias" in file1_lines[i]:
temp = temp.replace("bias", "beta")
if "running_mean" in file1_lines[i]:
temp = temp.replace("running_mean", "moving_mean")
if "running_var" in file1_lines[i]:
temp = temp.replace("running_var", "moving_variance")
mapping[file1_lines[i]] = temp
if temp not in file2_lines:
print(temp)
print("wrong")
with open('mapping.json', 'w') as json_file:
json.dump(mapping, json_file, indent=4)
print("Mapping saved to 'mapping.json'")
4 根据得到到mapping.json文件生成mindspore需要的ckpt文件;
核心:先载入一个pytorch的模型,用mindspore的key和pytorch的value
# convert pytorch model to mindspore
import json
import os
from torchvision.models import resnet50
import mindspore as ms
# load param_map json
with open("mapping.json", "r") as json_file:
param_name_map = json.load(json_file)
net = resnet50(pretrained=True)
params_dict = net.state_dict()
# conversion
ms_params = []
for name, value in params_dict.items():
each_param = dict()
if name not in param_name_map:
continue
each_param["name"] = param_name_map[name]
each_param["data"] = ms.Tensor(value.numpy())
ms_params.append(each_param)
ms.save_checkpoint(ms_params, "resnet50_ms.ckpt")