特别说明：参考官方开源的yolov7代码、瑞芯微官方文档、地平线的官方文档，如有侵权告知删，谢谢。
模型和完整仿真测试代码，放在github上参考链接 模型和代码。
1 模型和训练
训练代码参考官方开源的yolov7训练代码，由于SiLU在有些板端芯片上还不支持，因此将其改为ReLU。
2 导出 yolov7 onnx
后处理中有些算在板端芯片上效率低或者不支持，导出 onnx 需要将板端芯片不友好或不支持算子规避掉。导出onnx修改的部分。
规避板端芯片支持不友好的算子，修改yolo.py文件。（示例代码是yolov7开源的较早版本，若不一致对应修改即可）
yolo.py 的完整修改如下
import argparse
import logging
import sys
from copy import deepcopy
sys.path.append('./') # to run '$ python *.py' files in subdirectories
logger = logging.getLogger(__name__)
import torch
from models.common import *
from models.experimental import *
from utils.autoanchor import check_anchor_order
from utils.general import make_divisible, check_file, set_logging
from utils.torch_utils import time_synchronized, fuse_conv_and_bn, model_info, scale_img, initialize_weights, \
select_device, copy_attr
from utils.loss import SigmoidBin
try:
import thop # for FLOPS computation
except ImportError:
thop = None
class Detect(nn.Module):
stride = None # strides computed during build
export = False # onnx export
end2end = False
include_nms = False
concat = False
def __init__(self, nc=80, anchors=(), ch=()): # detection layer
super(Detect, self).__init__()
self.nc = nc # number of classes
self.no = nc + 5 # number of outputs per anchor
self.nl = len(anchors) # number of detection layers
self.na = len(anchors[0]) // 2 # number of anchors
self.grid = [torch.zeros(1)] * self.nl # init grid
a = torch.tensor(anchors).float().view(self.nl, -1, 2)
self.register_buffer('anchors', a) # shape(nl,na,2)
self.register_buffer('anchor_grid', a.clone().view(self.nl, 1, -1, 1, 1, 2)) # shape(nl,1,na,1,1,2)
self.m = nn.ModuleList(nn.Conv2d(x, self.no * self.na, 1) for x in ch) # output conv
def forward(self, x):
# x = x.copy() # for profiling
z = [] # inference output
self.training |= self.export
for i in range(self.nl):
x[i] = self.m[i](x[i]) # conv
z.append(x[i])
continue
bs, _, ny, nx = x[i].shape # x(bs,255,20,20) to x(bs,3,20,20,85)
x[i] = x[i].view(bs, self.na, self.no, ny, nx).permute(0, 1, 3, 4, 2).contiguous()
if not self.training: # inference
if self.grid[i].shape[2:4] != x[i].shape[2:4]:
self.grid[i] = self._make_grid(nx, ny).to(x[i].device)
y = x[i].sigmoid()
if not torch.onnx.is_in_onnx_export():
y[..., 0:2] = (y[..., 0:2] * 2. - 0.5 + self.grid[i]) * self.stride[i] # xy
y[..., 2:4] = (y[..., 2:4] * 2) ** 2 * self.anchor_grid[i] # wh
else:
xy, wh, conf = y.split((2, 2, self.nc + 1), 4) # y.tensor_split((2, 4, 5), 4) # torch 1.8.0
xy = xy * (2. * self.stride[i]) + (self.stride[i] * (self.grid[i] - 0.5)) # new xy
wh = wh ** 2 * (4 * self.anchor_grid[i].data) # new wh
y = torch.cat((xy, wh, conf), 4)
z.append(y.view(bs, -1, self.no))
return z
if self.training:
out = x
elif self.end2end:
out = torch.cat(z, 1)
elif self.include_nms:
z = self.convert(z)
out = (z,)
elif self.concat:
out = torch.cat(z, 1)
else:
out = (torch.cat(z, 1), x)
return out
@staticmethod
def _make_grid(nx=20, ny=20):
yv, xv = torch.meshgrid([torch.arange(ny), torch.arange(nx)])
return torch.stack((xv, yv), 2).view((1, 1, ny, nx, 2)).float()
def convert(self, z):
z = torch.cat(z, 1)
box = z[:, :, :4]
conf = z[:, :, 4:5]
score = z[:, :, 5:]
score *= conf
convert_matrix = torch.tensor([[1, 0, 1, 0], [0, 1, 0, 1], [-0.5, 0, 0.5, 0], [0, -0.5, 0, 0.5]],
dtype=torch.float32,
device=z.device)
box @= convert_matrix
return (box, score)
class IDetect(nn.Module):
stride = None # strides computed during build
export = False # onnx export
end2end = False
include_nms = False
concat = False
def __init__(self, nc=80, anchors=(), ch=()): # detection layer
super(IDetect, self).__init__()
self.nc = nc # number of classes
self.no = nc + 5 # number of outputs per anchor
self.nl = len(anchors) # number of detection layers
self.na = len(anchors[0]) // 2 # number of anchors
self.grid = [torch.zeros(1)] * self.nl # init grid
a = torch.tensor(anchors).float().view(self.nl, -1, 2)
self.register_buffer('anchors', a) # shape(nl,na,2)
self.register_buffer('anchor_grid', a.clone().view(self.nl, 1, -1, 1, 1, 2)) # shape(nl,1,na,1,1,2)
self.m = nn.ModuleList(nn.Conv2d(x, self.no * self.na, 1) for x in ch) # output conv
self.ia = nn.ModuleList(ImplicitA(x) for x in ch)
self.im = nn.ModuleList(ImplicitM(self.no * self.na) for _ in ch)
def forward(self, x):
# x = x.copy() # for profiling
z = [] # inference output
self.training |= self.export
for i in range(self.nl):
x[i] = self.m[i](self.ia[i](x[i])) # conv
x[i] = self.im[i](x[i])
z.append(x[i])
continue
bs, _, ny, nx = x[i].shape # x(bs,255,20,20) to x(bs,3,20,20,85)
x[i] = x[i].view(bs, self.na, self.no, ny, nx).permute(0, 1, 3, 4, 2).contiguous()
if not self.training: # inference
if self.grid[i].shape[2:4] != x[i].shape[2:4]:
self.grid[i] = self._make_grid(nx, ny).to(x[i].device)
y = x[i].sigmoid()
y[..., 0:2] = (y[..., 0:2] * 2. - 0.5 + self.grid[i]) * self.stride[i] # xy
y[..., 2:4] = (y[..., 2:4] * 2) ** 2 * self.anchor_grid[i] # wh
z.append(y.view(bs, -1, self.no))
return z
return x if self.training else (torch.cat(z, 1), x)
def fuseforward(self, x):
# x = x.copy() # for profiling
z = [] # inference output
self.training |= self.export
for i in range(self.nl):
x[i] = self.m[i](x[i]) # conv
# print(x[i])
z.append(x[i])
continue
bs, _, ny, nx = x[i].shape # x(bs,255,20,20) to x(bs,3,20,20,85)
x[i] = x[i].view(bs, self.na, self.no, ny, nx).permute(0, 1, 3, 4, 2).contiguous()
if not self.training: # inference
if self.grid[i].shape[2:4] != x[i].shape[2:4]:
self.grid[i] = self._make_grid(nx, ny).to(x[i].device)
y = x[i].sigmoid()
if not torch.onnx.is_in_onnx_export():
y[..., 0:2] = (y[..., 0:2] * 2. - 0.5 + self.grid[i]) * self.stride[i] # xy
y[..., 2:4] = (y[..., 2:4] * 2) ** 2 * self.anchor_grid[i] # wh
else:
xy, wh, conf = y.split((2, 2, self.nc + 1), 4) # y.tensor_split((2, 4, 5), 4) # torch 1.8.0
xy = xy * (2. * self.stride[i]) + (self.stride[i] * (self.grid[i] - 0.5)) # new xy
wh = wh ** 2 * (4 * self.anchor_grid[i].data) # new wh
y = torch.cat((xy, wh, conf), 4)
z.append(y.view(bs, -1, self.no))
return z
if self.training:
out = x
elif self.end2end:
out = torch.cat(z, 1)
elif self.include_nms:
z = self.convert(z)
out = (z,)
elif self.concat:
out = torch.cat(z, 1)
else:
out = (torch.cat(z, 1), x)
return out
def fuse(self):
print("IDetect.fuse")
# fuse ImplicitA and Convolution
for i in range(len(self.m)):
c1, c2, _, _ = self.m[i].weight.shape
c1_, c2_, _, _ = self.ia[i].implicit.shape
self.m[i].bias += torch.matmul(self.m[i].weight.reshape(c1, c2), self.ia[i].implicit.reshape(c2_, c1_)).squeeze(1)
# fuse ImplicitM and Convolution
for i in range(len(self.m)):
c1, c2, _, _ = self.im[i].implicit.shape
self.m[i].bias *= self.im[i].implicit.reshape(c2)
self.m[i].weight *= self.im[i].implicit.transpose(0, 1)
@staticmethod
def _make_grid(nx=20, ny=20):
yv, xv = torch.meshgrid([torch.arange(ny), torch.arange(nx)])
return torch.stack((xv, yv), 2).view((1, 1, ny, nx, 2)).float()
def convert(self, z):
print('convert ...')
z = torch.cat(z, 1)
box = z[:, :, :4]
conf = z[:, :, 4:5]
score = z[:, :, 5:]
score *= conf
convert_matrix = torch.tensor([[1, 0, 1, 0], [0, 1, 0, 1], [-0.5, 0, 0.5, 0], [0, -0.5, 0, 0.5]],
dtype=torch.float32,
device=z.device)
box @= convert_matrix
return (box, score)
class IKeypoint(nn.Module):
stride = None # strides computed during build
export = False # onnx export
def __init__(self, nc=80, anchors=(), nkpt=17, ch=(), inplace=True, dw_conv_kpt=False): # detection layer
super(IKeypoint, self).__init__()
self.nc = nc # number of classes
self.nkpt = nkpt
self.dw_conv_kpt = dw_conv_kpt
self.no_det = (nc + 5) # number of outputs per anchor for box and class
self.no_kpt = 3 * self.nkpt ## number of outputs per anchor for keypoints
self.no = self.no_det + self.no_kpt
self.nl = len(anchors) # number of detection layers
self.na = len(anchors[0]) // 2 # number of anchors
self.grid = [torch.zeros(1)] * self.nl # init grid
self.flip_test = False
a = torch.tensor(anchors).float().view(self.nl, -1, 2)
self.register_buffer('anchors', a) # shape(nl,na,2)
self.register_buffer('anchor_grid', a.clone().view(self.nl, 1, -1, 1, 1, 2)) # shape(nl,1,na,1,1,2)
self.m = nn.ModuleList(nn.Conv2d(x, self.no_det * self.na, 1) for x in ch) # output conv
self.ia = nn.ModuleList(ImplicitA(x) for x in ch)
self.im = nn.ModuleList(ImplicitM(self.no_det * self.na) for _ in ch)
if self.nkpt is not None:
if self.dw_conv_kpt: # keypoint head is slightly more complex
self.m_kpt = nn.ModuleList(
nn.Sequential(DWConv(x, x, k=3), Conv(x, x),
DWConv(x, x, k=3), Conv(x, x),
DWConv(x, x, k=3), Conv(x, x),
DWConv(x, x, k=3), Conv(x, x),
DWConv(x, x, k=3), Conv(x, x),
DWConv(x, x, k=3), nn.Conv2d(x, self.no_kpt * self.na, 1)) for x in ch)
else: # keypoint head is a single convolution
self.m_kpt = nn.ModuleList(nn.Conv2d(x, self.no_kpt * self.na, 1) for x in ch)
self.inplace = inplace # use in-place ops (e.g. slice assignment)
def forward(self, x):
# x = x.copy() # for profiling
z = [] # inference output
self.training |= self.export
for i in range(self.nl):
if self.nkpt is None or self.nkpt == 0:
x[i] = self.im[i](self.m[i](self.ia[i](x[i]))) # conv
else:
x[i] = torch.cat((self.im[i](self.m[i](self.ia[i](x[i]))), self.m_kpt[i](x[i])), axis=1)
bs, _, ny, nx = x[i].shape # x(bs,255,20,20) to x(bs,3,20,20,85)
x[i] = x[i].view(bs, self.na, self.no, ny, nx).permute(0, 1, 3, 4, 2).contiguous()
x_det = x[i][..., :6]
x_kpt = x[i][..., 6:]
if not self.training: # inference
if self.grid[i].shape[2:4] != x[i].shape[2:4]:
self.grid[i] = self._make_grid(nx, ny).to(x[i].device)
kpt_grid_x = self.grid[i][..., 0:1]
kpt_grid_y = self.grid[i][..., 1:2]
if self.nkpt == 0:
y = x[i].sigmoid()
else:
y = x_det.sigmoid()
if self.inplace:
xy = (y[..., 0:2] * 2. - 0.5 + self.grid[i]) * self.stride[i] # xy
wh = (y[..., 2:4] * 2) ** 2 * self.anchor_grid[i].view(1, self.na, 1, 1, 2) # wh
if self.nkpt != 0:
x_kpt[..., 0::3] = (x_kpt[..., ::3] * 2. - 0.5 + kpt_grid_x.repeat(1, 1, 1, 1, 17)) * \
self.stride[i] # xy
x_kpt[..., 1::3] = (x_kpt[..., 1::3] * 2. - 0.5 + kpt_grid_y.repeat(1, 1, 1, 1, 17)) * \
self.stride[i] # xy
# x_kpt[..., 0::3] = (x_kpt[..., ::3] + kpt_grid_x.repeat(1,1,1,1,17)) * self.stride[i] # xy
# x_kpt[..., 1::3] = (x_kpt[..., 1::3] + kpt_grid_y.repeat(1,1,1,1,17)) * self.stride[i] # xy
# print('=============')
# print(self.anchor_grid[i].shape)
# print(self.anchor_grid[i][...,0].unsqueeze(4).shape)
# print(x_kpt[..., 0::3].shape)
# x_kpt[..., 0::3] = ((x_kpt[..., 0::3].tanh() * 2.) ** 3 * self.anchor_grid[i][...,0].unsqueeze(4).repeat(1,1,1,1,self.nkpt)) + kpt_grid_x.repeat(1,1,1,1,17) * self.stride[i] # xy
# x_kpt[..., 1::3] = ((x_kpt[..., 1::3].tanh() * 2.) ** 3 * self.anchor_grid[i][...,1].unsqueeze(4).repeat(1,1,1,1,self.nkpt)) + kpt_grid_y.repeat(1,1,1,1,17) * self.stride[i] # xy
# x_kpt[..., 0::3] = (((x_kpt[..., 0::3].sigmoid() * 4.) ** 2 - 8.) * self.anchor_grid[i][...,0].unsqueeze(4).repeat(1,1,1,1,self.nkpt)) + kpt_grid_x.repeat(1,1,1,1,17) * self.stride[i] # xy
# x_kpt[..., 1::3] = (((x_kpt[..., 1::3].sigmoid() * 4.) ** 2 - 8.) * self.anchor_grid[i][...,1].unsqueeze(4).repeat(1,1,1,1,self.nkpt)) + kpt_grid_y.repeat(1,1,1,1,17) * self.stride[i] # xy
x_kpt[..., 2::3] = x_kpt[..., 2::3].sigmoid()
y = torch.cat((xy, wh, y[..., 4:], x_kpt), dim=-1)
else: # for YOLOv5 on AWS Inferentia https://github.com/ultralytics/yolov5/pull/2953
xy = (y[..., 0:2] * 2. - 0.5 + self.grid[i]) * self.stride[i] # xy
wh = (y[..., 2:4] * 2) ** 2 * self.anchor_grid[i] # wh
if self.nkpt != 0:
y[..., 6:] = (y[..., 6:] * 2. - 0.5 + self.grid[i].repeat((1, 1, 1, 1, self.nkpt))) * \
self.stride[i] # xy
y = torch.cat((xy, wh, y[..., 4:]), -1)
z.append(y.view(bs, -1, self.no))
return x if self.training else (torch.cat(z, 1), x)
@staticmethod
def _make_grid(nx=20, ny=20):
yv, xv = torch.meshgrid([torch.arange(ny), torch.arange(nx)])
return torch.stack((xv, yv), 2).view((1, 1, ny, nx, 2)).float()
class IAuxDetect(nn.Module):
stride = None # strides computed during build
export = False # onnx export
end2end = False
include_nms = False
concat = False
def __init__(self, nc=80, anchors=(), ch=()): # detection layer
super(IAuxDetect, self).__init__()
self.nc = nc # number of classes
self.no = nc + 5 # number of outputs per anchor
self.nl = len(anchors) # number of detection layers
self.na = len(anchors[0]) // 2 # number of anchors
self.grid = [torch.zeros(1)] * self.nl # init grid
a = torch.tensor(anchors).float().view(self.nl, -1, 2)
self.register_buffer('anchors', a) # shape(nl,na,2)
self.register_buffer('anchor_grid', a.clone().view(self.nl, 1, -1, 1, 1, 2)) # shape(nl,1,na,1,1,2)
self.m = nn.ModuleList(nn.Conv2d(x, self.no * self.na, 1) for x in ch[:self.nl]) # output conv
self.m2 = nn.ModuleList(nn.Conv2d(x, self.no * self.na, 1) for x in ch[self.nl:]) # output conv
self.ia = nn.ModuleList(ImplicitA(x) for x in ch[:self.nl])
self.im = nn.ModuleList(ImplicitM(self.no * self.na) for _ in ch[:self.nl])
def forward(self, x):
# x = x.copy() # for profiling
z = [] # inference output
self.training |= self.export
for i in range(self.nl):
x[i] = self.m[i](self.ia[i](x[i])) # conv
x[i] = self.im[i](x[i])
bs, _, ny, nx = x[i].shape # x(bs,255,20,20) to x(bs,3,20,20,85)
x[i] = x[i].view(bs, self.na, self.no, ny, nx).permute(0, 1, 3, 4, 2).contiguous()
x[i + self.nl] = self.m2[i](x[i + self.nl])
x[i + self.nl] = x[i + self.nl].view(bs, self.na, self.no, ny, nx).permute(0, 1, 3, 4, 2).contiguous()
if not self.training: # inference
if self.grid[i].shape[2:4] != x[i].shape[2:4]:
self.grid[i] = self._make_grid(nx, ny).to(x[i].device)
y = x[i].sigmoid()
if not torch.onnx.is_in_onnx_export():
y[..., 0:2] = (y[..., 0:2] * 2. - 0.5 + self.grid[i]) * self.stride[i] # xy
y[..., 2:4] = (y[..., 2:4] * 2) ** 2 * self.anchor_grid[i] # wh
else:
xy, wh, conf = y.split((2, 2, self.nc + 1), 4) # y.tensor_split((2, 4, 5), 4) # torch 1.8.0
xy = xy * (2. * self.stride[i]) + (self.stride[i] * (self.grid[i] - 0.5)) # new xy
wh = wh ** 2 * (4 * self.anchor_grid[i].data) # new wh
y = torch.cat((xy, wh, conf), 4)
z.append(y.view(bs, -1, self.no))
return x if self.training else (torch.cat(z, 1), x[:self.nl])
def fuseforward(self, x):
# x = x.copy() # for profiling
z = [] # inference output
self.training |= self.export
for i in range(self.nl):
x[i] = self.m[i](x[i]) # conv
bs, _, ny, nx = x[i].shape # x(bs,255,20,20) to x(bs,3,20,20,85)
x[i] = x[i].view(bs, self.na, self.no, ny, nx).permute(0, 1, 3, 4, 2).contiguous()
if not self.training: # inference
if self.grid[i].shape[2:4] != x[i].shape[2:4]:
self.grid[i] = self._make_grid(nx, ny).to(x[i].device)
y = x[i].sigmoid()
if not torch.onnx.is_in_onnx_export():
y[..., 0:2] = (y[..., 0:2] * 2. - 0.5 + self.grid[i]) * self.stride[i] # xy
y[..., 2:4] = (y[..., 2:4] * 2) ** 2 * self.anchor_grid[i] # wh
else:
xy = (y[..., 0:2] * 2. - 0.5 + self.grid[i]) * self.stride[i] # xy
wh = (y[..., 2:4] * 2) ** 2 * self.anchor_grid[i].data # wh
y = torch.cat((xy, wh, y[..., 4:]), -1)
z.append(y.view(bs, -1, self.no))
if self.training:
out = x
elif self.end2end:
out = torch.cat(z, 1)
elif self.include_nms:
z = self.convert(z)
out = (z,)
elif self.concat:
out = torch.cat(z, 1)
else:
out = (torch.cat(z, 1), x)
return out
def fuse(self):
print("IAuxDetect.fuse")
# fuse ImplicitA and Convolution
for i in range(len(self.m)):
c1, c2, _, _ = self.m[i].weight.shape
c1_, c2_, _, _ = self.ia[i].implicit.shape
self.m[i].bias += torch.matmul(self.m[i].weight.reshape(c1, c2),
self.ia[i].implicit.reshape(c2_, c1_)).squeeze(1)
# fuse ImplicitM and Convolution
for i in range(len(self.m)):
c1, c2, _, _ = self.im[i].implicit.shape
self.m[i].bias *= self.im[i].implicit.reshape(c2)
self.m[i].weight *= self.im[i].implicit.transpose(0, 1)
@staticmethod
def _make_grid(nx=20, ny=20):
yv, xv = torch.meshgrid([torch.arange(ny), torch.arange(nx)])
return torch.stack((xv, yv), 2).view((1, 1, ny, nx, 2)).float()
def convert(self, z):
z = torch.cat(z, 1)
box = z[:, :, :4]
conf = z[:, :, 4:5]
score = z[:, :, 5:]
score *= conf
convert_matrix = torch.tensor([[1, 0, 1, 0], [0, 1, 0, 1], [-0.5, 0, 0.5, 0], [0, -0.5, 0, 0.5]],
dtype=torch.float32,
device=z.device)
box @= convert_matrix
return (box, score)
class IBin(nn.Module):
stride = None # strides computed during build
export = False # onnx export
def __init__(self, nc=80, anchors=(), ch=(), bin_count=21): # detection layer
super(IBin, self).__init__()
self.nc = nc # number of classes
self.bin_count = bin_count
self.w_bin_sigmoid = SigmoidBin(bin_count=self.bin_count, min=0.0, max=4.0)
self.h_bin_sigmoid = SigmoidBin(bin_count=self.bin_count, min=0.0, max=4.0)
# classes, x,y,obj
self.no = nc + 3 + \
self.w_bin_sigmoid.get_length() + self.h_bin_sigmoid.get_length() # w-bce, h-bce
# + self.x_bin_sigmoid.get_length() + self.y_bin_sigmoid.get_length()
self.nl = len(anchors) # number of detection layers
self.na = len(anchors[0]) // 2 # number of anchors
self.grid = [torch.zeros(1)] * self.nl # init grid
a = torch.tensor(anchors).float().view(self.nl, -1, 2)
self.register_buffer('anchors', a) # shape(nl,na,2)
self.register_buffer('anchor_grid', a.clone().view(self.nl, 1, -1, 1, 1, 2)) # shape(nl,1,na,1,1,2)
self.m = nn.ModuleList(nn.Conv2d(x, self.no * self.na, 1) for x in ch) # output conv
self.ia = nn.ModuleList(ImplicitA(x) for x in ch)
self.im = nn.ModuleList(ImplicitM(self.no * self.na) for _ in ch)
def forward(self, x):
# self.x_bin_sigmoid.use_fw_regression = True
# self.y_bin_sigmoid.use_fw_regression = True
self.w_bin_sigmoid.use_fw_regression = True
self.h_bin_sigmoid.use_fw_regression = True
# x = x.copy() # for profiling
z = [] # inference output
self.training |= self.export
for i in range(self.nl):
x[i] = self.m[i](self.ia[i](x[i])) # conv
x[i] = self.im[i](x[i])
bs, _, ny, nx = x[i].shape # x(bs,255,20,20) to x(bs,3,20,20,85)
x[i] = x[i].view(bs, self.na, self.no, ny, nx).permute(0, 1, 3, 4, 2).contiguous()
if not self.training: # inference
if self.grid[i].shape[2:4] != x[i].shape[2:4]:
self.grid[i] = self._make_grid(nx, ny).to(x[i].device)
y = x[i].sigmoid()
y[..., 0:2] = (y[..., 0:2] * 2. - 0.5 + self.grid[i]) * self.stride[i] # xy
# y[..., 2:4] = (y[..., 2:4] * 2) ** 2 * self.anchor_grid[i] # wh
# px = (self.x_bin_sigmoid.forward(y[..., 0:12]) + self.grid[i][..., 0]) * self.stride[i]
# py = (self.y_bin_sigmoid.forward(y[..., 12:24]) + self.grid[i][..., 1]) * self.stride[i]
pw = self.w_bin_sigmoid.forward(y[..., 2:24]) * self.anchor_grid[i][..., 0]
ph = self.h_bin_sigmoid.forward(y[..., 24:46]) * self.anchor_grid[i][..., 1]
# y[..., 0] = px
# y[..., 1] = py
y[..., 2] = pw
y[..., 3] = ph
y = torch.cat((y[..., 0:4], y[..., 46:]), dim=-1)
z.append(y.view(bs, -1, y.shape[-1]))
return x if self.training else (torch.cat(z, 1), x)
@staticmethod
def _make_grid(nx=20, ny=20):
yv, xv = torch.meshgrid([torch.arange(ny), torch.arange(nx)])
return torch.stack((xv, yv), 2).view((1, 1, ny, nx, 2)).float()
class Model(nn.Module):
def __init__(self, cfg='yolor-csp-c.yaml', ch=3, nc=None, anchors=None): # model, input channels, number of classes
super(Model, self).__init__()
self.traced = False
if isinstance(cfg, dict):
self.yaml = cfg # model dict
else: # is *.yaml
import yaml # for torch hub
self.yaml_file = Path(cfg).name
with open(cfg) as f:
self.yaml = yaml.load(f, Loader=yaml.SafeLoader) # model dict
# Define model
ch = self.yaml['ch'] = self.yaml.get('ch', ch) # input channels
if nc and nc != self.yaml['nc']:
logger.info(f"Overriding model.yaml nc={
self.yaml['nc']} with nc={
nc}")
self.yaml['nc'] = nc # override yaml value
if anchors:
logger.info(f'Overriding model.yaml anchors with anchors={
anchors}')
self.yaml['anchors'] = round(anchors) # override yaml value
self.model, self.save = parse_model(deepcopy(self.yaml), ch=[ch]) # model, savelist
self.names = [str(i) for i in range(self.yaml['nc'])] # default names
# print([x.shape for x in self.forward(torch.zeros(1, ch, 64, 64))])
# Build strides, anchors
m = self.model[-1] # Detect()
if isinstance(m, Detect):
s = 256 # 2x min stride
m.stride = torch.tensor([s / x.shape[-2] for x in self.forward(torch.zeros(1, ch, s, s))]) # forward
check_anchor_order(m)
m.anchors /= m.stride.view(-1, 1, 1)
self.stride = m.stride
self._initialize_biases() # only run once
# print('Strides: %s' % m.stride.tolist())
if isinstance(m, IDetect):
s = 256 # 2x min stride
m.stride = torch.tensor([s / x.shape[-2] for x in self.forward(torch.zeros(1, ch, s, s))]) # forward
check_anchor_order(m)
m.anchors /= m.stride.view(-1, 1, 1)
self.stride = m.stride
self._initialize_biases() # only run once
# print('Strides: %s' % m.stride.tolist())
if isinstance(m, IAuxDetect):
s = 256 # 2x min stride
m.stride = torch.tensor([s / x.shape[-2] for x in self.forward(torch.zeros(1, ch, s, s))[:4]]) # forward
# print(m.stride)
check_anchor_order(m)
m.anchors /= m.stride.view(-1, 1, 1)
self.stride = m.stride
self._initialize_aux_biases() # only run once
# print('Strides: %s' % m.stride.tolist())
if isinstance(m, IBin):
s = 256 # 2x min stride
m.stride = torch.tensor([s / x.shape[-2] for x in self.forward(torch.zeros(1, ch, s, s))]) # forward
check_anchor_order(m)
m.anchors /= m.stride.view(-1, 1, 1)
self.stride = m.stride
self._initialize_biases_bin() # only run once
# print('Strides: %s' % m.stride.tolist())
if isinstance(m, IKeypoint):
s = 256 # 2x min stride
m.stride = torch.tensor([s / x.shape[-2] for x in self.forward(torch.zeros(1, ch, s, s))]) # forward
check_anchor_order(m)
m.anchors /= m.stride.view(-1, 1, 1)
self.stride = m.stride
self._initialize_biases_kpt() # only run once
# print('Strides: %s' % m.stride.tolist())
# Init weights, biases
initialize_weights(self)
self.info()
logger.info('')
def forward(self, x, augment=False, profile=False):
if augment:
img_size = x.shape[-2:] # height, width
s = [1, 0.83, 0.67] # scales
f = [None, 3, None] # flips (2-ud, 3-lr)
y = [] # outputs
for si, fi in zip(s, f):
xi = scale_img(x.flip(fi) if fi else x, si, gs=int(self.stride.max()))
yi = self.forward_once(xi)[0] # forward
# cv2.imwrite(f'img_{si}.jpg', 255 * xi[0].cpu().numpy().transpose((1, 2, 0))[:, :, ::-1]) # save
yi[..., :4] /= si # de-scale
if fi == 2:
yi[..., 1] = img_size[0] - yi[..., 1] # de-flip ud
elif fi == 3:
yi[..., 0] = img_size[1] - yi[..., 0] # de-flip lr
y.append(yi)
return torch.cat(y, 1), None # augmented inference, train
else:
return self.forward_once(x, profile) # single-scale inference, train
def forward_once(self, x, profile=False):
y, dt = [], [] # outputs
for m in self.model:
if m.f != -1: # if not from previous layer
x = y[m.f] if isinstance(m.f, int) else [x if j == -1 else y[j] for j in m.f] # from earlier layers
if not hasattr(self, 'traced'):
self.traced = False
if self.traced:
if isinstance(m, Detect) or isinstance(m, IDetect) or isinstance(m, IAuxDetect) or isinstance(m,
IKeypoint):
break
if profile:
c = isinstance(m, (Detect, IDetect, IAuxDetect, IBin))
o = thop.profile(m, inputs=(x.copy() if c else x,), verbose=False)[0] / 1E9 * 2 if thop else 0 # FLOPS
for _ in range(10):
m(x.copy() if c else x)
t = time_synchronized()
for _ in range(10):
m(x.copy() if c else x)
dt.append((time_synchronized() - t) * 100)
print('%10.1f%10.0f%10.1fms %-40s' % (o, m.np, dt[-1], m.type))
x = m(x) # run
y.append(x if m.i in self.save else None) # save output
if profile:
print('%.1fms total' % sum(dt))
return x
def _initialize_biases(self, cf=None): # initialize biases into Detect(), cf is class frequency
# https://arxiv.org/abs/1708.02002 section 3.3
# cf = torch.bincount(torch.tensor(np.concatenate(dataset.labels, 0)[:, 0]).long(), minlength=nc) + 1.
m = self.model[-1] # Detect() module
for mi, s in zip(m.m, m.stride): # from
b = mi.bias.view(m.na, -1) # conv.bias(255) to (3,85)
b.data[:, 4] += math.log(8 / (640 / s) ** 2) # obj (8 objects per 640 image)
b.data[:, 5:] += math.log(0.6 / (m.nc - 0.99)) if cf is None else torch.log(cf / cf.sum()) # cls
mi.bias = torch.nn.Parameter(b.view(-1), requires_grad=True)
def _initialize_aux_biases(self, cf=None): # initialize biases into Detect(), cf is class frequency
# https://arxiv.org/abs/1708.02002 section 3.3
# cf = torch.bincount(torch.tensor(np.concatenate(dataset.labels, 0)[:, 0]).long(), minlength=nc) + 1.
m = self.model[-1] # Detect() module
for mi, mi2, s in zip(m.m, m.m2, m.stride): # from
b = mi.bias.view(m.na, -1) # conv.bias(255) to (3,85)
b.data[:, 4] += math.log(8 / (640 / s) ** 2) # obj (8 objects per 640 image)
b.data[:, 5:] += math.log(0.6 / (m.nc - 0.99)) if cf is None else torch.log(cf / cf.sum()) # cls
mi.bias = torch.nn.Parameter(b.view(-1), requires_grad=True)
b2 = mi2.bias.view(m.na, -1) # conv.bias(255) to (3,85)
b2.data[:, 4] += math.log(8 / (640 / s) ** 2) # obj (8 objects per 640 image)
b2.data[:, 5:] += math.log(0.6 / (m.nc - 0.99)) if cf is None else torch.log(cf / cf.sum()) # cls
mi2.bias = torch.nn.Parameter(b2.view(-1), requires_grad=True)
def _initialize_biases_bin(self, cf=None): # initialize biases into Detect(), cf is class frequency
# https://arxiv.org/abs/1708.02002 section 3.3
# cf = torch.bincount(torch.tensor(np.concatenate(dataset.labels, 0)[:, 0]).long(), minlength=nc) + 1.
m = self.model[-1] # Bin() module
bc = m.bin_count
for mi, s in zip(m.m, m.stride): # from
b = mi.bias.view(m.na, -1) # conv.bias(255) to (3,85)
old = b[:, (0, 1, 2, bc + 3)].data
obj_idx = 2 * bc + 4
b[:, :obj_idx].data += math.log(0.6 / (bc + 1 - 0.99))
b[:, obj_idx].data += math.log(8 / (640 / s) ** 2) # obj (8 objects per 640 image)
b[:, (obj_idx + 1):].data += math.log(0.6 / (m.nc - 0.99)) if cf is None else torch.log(
cf / cf.sum()) # cls
b[:, (0, 1, 2, bc + 3)].data = old
mi.bias = torch.nn.Parameter(b.view(-1), requires_grad=True)
def _initialize_biases_kpt(self, cf=None): # initialize biases into Detect(), cf is class frequency
# https://arxiv.org/abs/1708.02002 section 3.3
# cf = torch.bincount(torch.tensor(np.concatenate(dataset.labels, 0)[:, 0]).long(), minlength=nc) + 1.
m = self.model[-1] # Detect() module
for mi, s in zip(m.m, m.stride): # from
b = mi.bias.view(m.na, -1) # conv.bias(255) to (3,85)
b.data[:, 4] += math.log(8 / (640 / s) ** 2) # obj (8 objects per 640 image)
b.data[:, 5:] += math.log(0.6 / (m.nc - 0.99)) if cf is None else torch.log(cf / cf.sum()) # cls
mi.bias = torch.nn.Parameter(b.view(-1), requires_grad=True)
def _print_biases(self):
m = self.model[-1] # Detect() module
for mi in m.m: # from
b = mi.bias.detach().view(m.na, -1).T # conv.bias(255) to (3,85)
print(('%6g Conv2d.bias:' + '%10.3g' * 6) % (mi.weight.shape[1], *b[:5].mean(1).tolist(), b[5:].mean()))
# def _print_weights(self):
# for m in self.model.modules():
# if type(m) is Bottleneck:
# print('%10.3g' % (m.w.detach().sigmoid() * 2)) # shortcut weights
def fuse(self): # fuse model Conv2d() + BatchNorm2d() layers
print('Fusing layers... ')
for m in self.model.modules():
if isinstance(m, RepConv):
# print(f" fuse_repvgg_block")
m.fuse_repvgg_block()
elif isinstance(m, RepConv_OREPA):
# print(f" switch_to_deploy")
m.switch_to_deploy()
elif type(m) is Conv and hasattr(m, 'bn'):
m.conv = fuse_conv_and_bn(m.conv, m.bn) # update conv
delattr(m, 'bn') # remove batchnorm
m.forward = m.fuseforward # update forward
elif isinstance(m, (IDetect, IAuxDetect)):
m.fuse()
m.forward = m.fuseforward
self.info()
return self
def nms(self, mode=True): # add or remove NMS module
present = type(self.model[-1]) is NMS # last layer is NMS
if mode and not present:
print('Adding NMS... ')
m = NMS() # module
m.f = -1 # from
m.i = self.model[-1].i + 1 # index
self.model.add_module(name='%s' % m.i, module=m) # add
self.eval()
elif not mode and present:
print('Removing NMS... ')
self.model = self.model[:-1] # remove
return self
def autoshape(self): # add autoShape module
print('Adding autoShape... ')
m = autoShape(self) # wrap model
copy_attr(m, self, include=('yaml', 'nc', 'hyp', 'names', 'stride'), exclude=()) # copy attributes
return m
def info(self, verbose=False, img_size=640): # print model information
model_info(self, verbose, img_size)
def parse_model(d, ch): # model_dict, input_channels(3)
logger.info('\n%3s%18s%3s%10s %-40s%-30s' % ('', 'from', 'n', 'params', 'module', 'arguments'))
anchors, nc, gd, gw = d['anchors'], d['nc'], d['depth_multiple'], d['width_multiple']
na = (len(anchors[0]) // 2) if isinstance(anchors, list) else anchors # number of anchors
no = na * (nc + 5) # number of outputs = anchors * (classes + 5)
layers, save, c2 = [], [], ch[-1] # layers, savelist, ch out
for i, (f, n, m, args) in enumerate(d['backbone'] + d['head']): # from, number, module, args
m = eval(m) if isinstance(m, str) else m # eval strings
for j, a in enumerate(args):
try:
args[j] = eval(a) if isinstance(a, str) else a # eval strings
except:
pass
n = max(round(n * gd), 1) if n > 1 else n # depth gain
if m in [nn.Conv2d, Conv, RobustConv, RobustConv2, DWConv, GhostConv, RepConv, RepConv_OREPA, DownC,
SPP, SPPF, SPPCSPC, GhostSPPCSPC, MixConv2d, Focus, Stem, GhostStem, CrossConv,
Bottleneck, BottleneckCSPA, BottleneckCSPB, BottleneckCSPC,
RepBottleneck, RepBottleneckCSPA, RepBottleneckCSPB, RepBottleneckCSPC,
Res, ResCSPA, ResCSPB, ResCSPC,
RepRes, RepResCSPA, RepResCSPB, RepResCSPC,
ResX, ResXCSPA, ResXCSPB, ResXCSPC,
RepResX, RepResXCSPA, RepResXCSPB, RepResXCSPC,
Ghost, GhostCSPA, GhostCSPB, GhostCSPC,
SwinTransformerBlock, STCSPA, STCSPB, STCSPC,
SwinTransformer2Block, ST2CSPA, ST2CSPB, ST2CSPC]:
c1, c2 = ch[f], args[0]
if c2 != no: # if not output
c2 = make_divisible(c2 * gw, 8)
args = [c1, c2, *args[1:]]
if m in [DownC, SPPCSPC, GhostSPPCSPC,
BottleneckCSPA, BottleneckCSPB, BottleneckCSPC,
RepBottleneckCSPA, RepBottleneckCSPB, RepBottleneckCSPC,
ResCSPA, ResCSPB, ResCSPC,
RepResCSPA, RepResCSPB, RepResCSPC,
ResXCSPA, ResXCSPB, ResXCSPC,
RepResXCSPA, RepResXCSPB, RepResXCSPC,
GhostCSPA, GhostCSPB, GhostCSPC,
STCSPA, STCSPB, STCSPC,
ST2CSPA, ST2CSPB, ST2CSPC]:
args.insert(2, n) # number of repeats
n = 1
elif m is nn.BatchNorm2d:
args = [ch[f]]
elif m is Concat:
c2 = sum([ch[x] for x in f])
elif m is Chuncat:
c2 = sum([ch[x] for x in f])
elif m is Shortcut:
c2 = ch[f[0]]
elif m is Foldcut:
c2 = ch[f] // 2
elif m in [Detect, IDetect, IAuxDetect, IBin, IKeypoint]:
args.append([ch[x] for x in f])
if isinstance(args[1], int): # number of anchors
args[1] = [list(range(args[1] * 2))] * len(f)
elif m is ReOrg:
c2 = ch[f] * 4
elif m is Contract:
c2 = ch[f] * args[0] ** 2
elif m is Expand:
c2 = ch[f] // args[0] ** 2
else:
c2 = ch[f]
m_ = nn.Sequential(*[m(*args) for _ in range(n)]) if n > 1 else m(*args) # module
t = str(m)[8:-2].replace('__main__.', '') # module type
np = sum([x.numel() for x in m_.parameters()]) # number params
m_.i, m_.f, m_.type, m_.np = i, f, t, np # attach index, 'from' index, type, number params
logger.info('%3s%18s%3s%10.0f %-40s%-30s' % (i, f, n, np, t, args)) # print
save.extend(x % i for x in ([f] if isinstance(f, int) else f) if x != -1) # append to savelist
layers.append(m_)
if i == 0:
ch = []
ch.append(c2)
return nn.Sequential(*layers), sorted(save)
if __name__ == '__main__':
parser = argparse.ArgumentParser()
parser.add_argument('--cfg', type=str, default='./cfg/training/yolov7.yaml', help='model.yaml')
parser.add_argument('--device', default='cpu', help='cuda device, i.e. 0 or 0,1,2,3 or cpu')
parser.add_argument('--profile', action='store_true', help='profile model speed')
opt = parser.parse_args()
opt.cfg = check_file(opt.cfg) # check file
set_logging()
device = select_device(opt.device)
# Create model
model = Model(opt.cfg).to(device)
model.train()
if opt.profile:
img = torch.rand(1, 3, 640, 640).to(device)
y = model(img, profile=False)
# Profile
# img = torch.rand(8 if torch.cuda.is_available() else 1, 3, 640, 640).to(device)
# y = model(img, profile=True)
# Tensorboard
# from torch.utils.tensorboard import SummaryWriter
# tb_writer = SummaryWriter()
# print("Run 'tensorboard --logdir=models/runs' to view tensorboard at http://localhost:6006/")
# tb_writer.add_graph(model.model, img) # add model to tensorboard
# tb_writer.add_image('test', img[0], dataformats='CWH') # add model to tensorboard
导出onnx模型，运行以下代码。
导出onnx代码。
import torch
import torch.onnx
from models.experimental import attempt_load
def export_onnx():
# Load model
model = attempt_load('./weights/best_yolov7_tiny.pt', map_location='cpu')
dummy_input = torch.ones(1, 3, 480, 640)
name = './weights/yolov7.onnx'
input_names = ["data"]
output_names = ["output1", "output2", "output3"]
torch.onnx.export(model.eval(), dummy_input, name, verbose=True, input_names=input_names, output_names=output_names, opset_version=10)
print("======================== convert onnx Finished! .... ")
if __name__ == '__main__':
export_onnx()
3 yolov7 onnx 测试效果
onnx模型和测试完整代码，放在github上代码。
注：图片来源网络，如果侵权告知删除。
4 yolov7 导出瑞芯微rknn和地平线horizon仿真测试
4.1 瑞芯微 rknn 仿真
瑞芯微环境搭建和详细步骤参考上一篇 【瑞芯微RKNN模型转换和PC端仿真】。
yolov7导出rknn模型代码和后处理参考 yolov7_rknn
4.2 地平线仿真
地平线环境搭建和详细步骤参考上一篇 【地平线Horizon模型转换和PC端仿真测试】。
yolov7导出地平线模型代码和后处理参考 yolov7_horizon