How to get the final result by feature graph?

[alter-xp]

❔Question

Hi, I am tring to inference yolov3.onnx in a special device, which does not support some operators when I export onnx with model.model[-1].export = False. so I have to use feature map to get the final result. I'm not sure how the three outputs [1,3,13,13,85], [1,3,26,26,85], [1,3,52,52,85] are converted into [110647,85]. Is there any example or tutorial of this?

[github-actions]

👋 Hello @alter-xp, thank you for your interest in 🚀 YOLOv5! Please visit our ⭐️ Tutorials to get started, where you can find quickstart guides for simple tasks like Custom Data Training all the way to advanced concepts like Hyperparameter Evolution.

If this is a 🐛 Bug Report, please provide screenshots and minimum viable code to reproduce your issue, otherwise we can not help you.

If this is a custom training ❓ Question, please provide as much information as possible, including dataset images, training logs, screenshots, and a public link to online W&B logging if available.

For business inquiries or professional support requests please visit https://www.ultralytics.com or email Glenn Jocher at glenn.jocher@ultralytics.com.

Requirements

Python 3.8 or later with all requirements.txt dependencies installed, including torch>=1.7. To install run:

$ pip install -r requirements.txt

Environments

YOLOv5 may be run in any of the following up-to-date verified environments (with all dependencies including CUDA/CUDNN, Python and PyTorch preinstalled):

• Google Colab and Kaggle notebooks with free GPU:
• Google Cloud Deep Learning VM. See GCP Quickstart Guide
• Amazon Deep Learning AMI. See AWS Quickstart Guide
• Docker Image. See Docker Quickstart Guide

Status

If this badge is green, all YOLOv5 GitHub Actions Continuous Integration (CI) tests are currently passing. CI tests verify correct operation of YOLOv5 training (train.py), testing (test.py), inference (detect.py) and export (export.py) on MacOS, Windows, and Ubuntu every 24 hours and on every commit.

[alter-xp]

Oh, I see

[glenn-jocher]

@alter-xp the Detect() layer takes in the 3 feature maps and outputs detections here:

yolov5/models/yolo.py

Line 22 in a3ecf0f

class Detect(nn.Module):

You'll probably need to include it in your ONNX export by setting export = False.

[alter-xp]

thanks a lot.

[CE-Noob]

哦，我明白了

I have also encountered this problem. May I ask how did you solve this problem?

[glenn-jocher]

@alter-xp @CE-Noob you can use the new --grid argument to export ONNX models with grid included in a single output:

python export.py --weights yolov5s.pt --grid

[CE-Noob]

@glenn-jocher This command will set model.model[-1].export=False. However,cv2.readNetFrom(model.onnx) failed when I export onnx with model.model[-1].export = False.In the actual deployment process, I have to use OpenCV. Before, I used this( https://github.com/hpc203/yolov5-dnn-cpp-python-v2) to transform the model so that onnx_model will done in Opencv. But now, I want to modify your code so that the three output(out0:[1,3,80,80,85],out1:[1,3,40,40,85],out2:[1,3,20,20,85]) could channge to just one output(shape:[25200,85]). Because of Opencv does not support slice layer,so I had modify the code in common.py in line 106
`
class Focus(nn.Module):
# Focus wh information into c-space
def init(self, c1, c2, k=1, s=1, p=None, g=1, act=True): # ch_in, ch_out, kernel, stride, padding, groups
super(Focus, self).init()
self.conv = Conv(c1 * 4, c2, k, s, p, g, act)
#self.contract = Contract(gain=2)

  def forward(self, x):  # x(b,c,w,h) -> y(b,4c,w/2,h/2)
      #return self.conv(torch.cat([x[..., ::2, ::2], x[..., 1::2, ::2], x[..., ::2, 1::2], x[..., 1::2, 1::2]], 1))
      N, C, H, W = x.size()  # assert (H / s == 0) and (W / s == 0), 'Indivisible gain'
      s = 2
      x = x.view(N, C, H // s, s, W // s, s)  # x(1,64,40,2,40,2)
      x = x.permute(0, 3, 5, 1, 2, 4).contiguous()  # x(1,2,2,64,40,40)
      yy=x.view(N, C * s * s, H // s, W // s)  # x(1,256,40,40)
      return self.conv(yy)

`
in the yolo.py line22:

`
class Detect(nn.Module):
stride = None # strides computed during build
export = False # onnx export

  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
          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()#(bs,3,20,20,85)
          #x[i] = x[i].view( self.na, self.no, ny, nx).permute(0, 2, 3, 1).contiguous()  # (bs,3,20,20,85)


          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
              #print(y.shape)
              z.append(y.view( -1, self.no).contiguous())
              #return torch.cat(z, 0)
          else:
              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( -1, self.no).contiguous())
      return (torch.cat(z, 0))

`
But it didn't work the way I wanted to,even though its output turned out to be shape [25200,85]. The slice layer still in the model

[CE-Noob]

@glenn-jocher thanks！I already know how to modify it. I can get the correct result. Thank you very much！

[anas-899]

@CE-Noob can you please share the steps you did until you get a correct inference with opencv DNN?

I am also getting the output (1x25200x85) as expected from opencv only if I disabled the following lines in yolo.py: (otherwise an exception is rising)
#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
how did you handle the slice operation within these lines, as one of the main problems in opencv is slice operation??