Maybe a bug? about pytorch-opcounter HOT 4 CLOSED

    flop = (input_size[0] / m.stride[0] * input_size[1] / m.stride[1]) * m.kernel_size[0] ** 2 * ((m.in_channels/m.groups) * (m.out_channels/m.groups) * m.groups)

     out_w * out_h * kw * kh * cin / groups * cout / groups * groups 
     => out_w * out_h * cout * kw * kh * cin / groups 
     => output_elements * kw * kh * cin / groups 
     => output_elements * ops_per_element * cin / groups 

from EXTD_64 import build_extd
from thop import profile
from thop import clever_format


num_classes = 2
img_dim = 640 
s3fd_net = build_extd('train', num_classes) # This is the [EXTD-Pytorch](https://github.com/clovaai/EXTD_Pytorch)
net = s3fd_net


input = torch.randn(1, 3, img_dim, img_dim)
flops, params = profile(net, inputs=(input, ))
flops, params = clever_format([flops, params], "%.8f")
print(flops,params)

1.08439040G 162.35200000K

import torch
import numpy as np
import torch.nn as nn


def compute_flops(model, image_size):
  flops = 0.
  input_size = image_size
  for m in model.modules():
    if len(list(m.children())) > 0:  # skip for non-leaf module
      continue
   
    if isinstance(m, nn.AvgPool2d) or isinstance(m, nn.MaxPool2d):
      input_size = input_size / 2.
    if isinstance(m, nn.Conv2d):
      if m.groups == 1:
        flop = (input_size[0] / m.stride[0] * input_size[1] / m.stride[1]) * m.kernel_size[0] ** 2 * m.in_channels * m.out_channels
      else:
        flop = (input_size[0] / m.stride[0] * input_size[1] / m.stride[1]) * m.kernel_size[0] ** 2 * ((m.in_channels/m.groups) * (m.out_channels/m.groups) * m.groups)
      flops += flop
      if m.stride[0] == 2: input_size = input_size / 2.
    print(m, flops, input_size)

  return flops / 1000000000., flops / 1000000



num_classes = 2
img_dim = 640 
s3fd_net = build_extd('train', num_classes) 
net = s3fd_net


gflops, mflops = compute_flops(net, np.array([img_dim, img_dim]))
print('# of params in Classification model: %d, flops: %.2f GFLOPS, %.2f MFLOPS, image_size: %d' % \
      (sum([p.data.nelement() for p in net.parameters()]), gflops, mflops, img_dim))

# of params in Classification model: 162352, flops: 11.15 GFLOPS, 11152.59 MFLOPS, image_size: 640

    m.total_ops = torch.Tensor([int(total_ops)])

    m.total_ops += torch.Tensor([int(total_ops)])