You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.

198 lines
8.4KB

  1. import argparse
  2. import yaml
  3. from models.common import *
  4. class Detect(nn.Module):
  5. def __init__(self, nc=80, anchors=()): # detection layer
  6. super(Detect, self).__init__()
  7. self.stride = None # strides computed during build
  8. self.nc = nc # number of classes
  9. self.no = nc + 5 # number of outputs per anchor
  10. self.nl = len(anchors) # number of detection layers
  11. self.na = len(anchors[0]) // 2 # number of anchors
  12. self.grid = [torch.zeros(1)] * self.nl # init grid
  13. a = torch.tensor(anchors).float().view(self.nl, -1, 2)
  14. self.register_buffer('anchors', a) # shape(nl,na,2)
  15. self.register_buffer('anchor_grid', a.clone().view(self.nl, 1, -1, 1, 1, 2)) # shape(nl,1,na,1,1,2)
  16. self.export = False # onnx export
  17. def forward(self, x):
  18. x = x.copy()
  19. z = [] # inference output
  20. self.training |= self.export
  21. for i in range(self.nl):
  22. bs, _, ny, nx = x[i].shape # x(bs,255,20,20) to x(bs,3,20,20,85)
  23. x[i] = x[i].view(bs, self.na, self.no, ny, nx).permute(0, 1, 3, 4, 2).contiguous()
  24. if not self.training: # inference
  25. if self.grid[i].shape[2:4] != x[i].shape[2:4]:
  26. self.grid[i] = self._make_grid(nx, ny).to(x[i].device)
  27. y = x[i].sigmoid()
  28. y[..., 0:2] = (y[..., 0:2] * 2. - 0.5 + self.grid[i]) * self.stride[i] # xy
  29. y[..., 2:4] = (y[..., 2:4] * 2) ** 2 * self.anchor_grid[i] # wh
  30. z.append(y.view(bs, -1, self.no))
  31. return x if self.training else (torch.cat(z, 1), x)
  32. @staticmethod
  33. def _make_grid(nx=20, ny=20):
  34. yv, xv = torch.meshgrid([torch.arange(ny), torch.arange(nx)])
  35. return torch.stack((xv, yv), 2).view((1, 1, ny, nx, 2)).float()
  36. class Model(nn.Module):
  37. def __init__(self, model_yaml='yolov5s.yaml'): # cfg, number of classes, depth-width gains
  38. super(Model, self).__init__()
  39. with open(model_yaml) as f:
  40. self.md = yaml.load(f, Loader=yaml.FullLoader) # model dict
  41. # Define model
  42. self.model, self.save, ch = parse_model(self.md, ch=[3]) # model, savelist, ch_out
  43. # print([x.shape for x in self.forward(torch.zeros(1, 3, 64, 64))])
  44. # Build strides, anchors
  45. m = self.model[-1] # Detect()
  46. m.stride = torch.tensor([64 / x.shape[-2] for x in self.forward(torch.zeros(1, 3, 64, 64))]) # forward
  47. m.anchors /= m.stride.view(-1, 1, 1)
  48. self.stride = m.stride
  49. # Init weights, biases
  50. torch_utils.initialize_weights(self)
  51. self._initialize_biases() # only run once
  52. torch_utils.model_info(self)
  53. print('')
  54. def forward(self, x, augment=False, profile=False):
  55. y, ts = [], 0 # outputs
  56. for m in self.model:
  57. if m.f != -1: # if not from previous layer
  58. 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
  59. if profile:
  60. import thop
  61. o = thop.profile(m, inputs=(x,), verbose=False)[0] / 1E9 * 2 # FLOPS
  62. t = torch_utils.time_synchronized()
  63. for _ in range(10):
  64. _ = m(x)
  65. dt = torch_utils.time_synchronized() - t
  66. ts += dt
  67. print('%10.1f%10.0f%10.1fms %-40s' % (o, m.np, dt * 100, m.type))
  68. x = m(x) # run
  69. y.append(x if m.i in self.save else None) # save output
  70. if profile:
  71. print(ts * 100)
  72. return x
  73. def _initialize_biases(self, cf=None): # initialize biases into Detect(), cf is class frequency
  74. # cf = torch.bincount(torch.tensor(np.concatenate(dataset.labels, 0)[:, 0]).long(), minlength=nc) + 1.
  75. m = self.model[-1] # Detect() module
  76. for f, s in zip(m.f, m.stride): #  from
  77. mi = self.model[f % m.i]
  78. b = mi.bias.view(m.na, -1) # conv.bias(255) to (3,85)
  79. b[:, 4] += math.log(8 / (640 / s) ** 2) # obj (8 objects per 640 image)
  80. b[:, 5:] += math.log(0.6 / (m.nc - 0.99)) if cf is None else torch.log(cf / cf.sum()) # cls
  81. mi.bias = torch.nn.Parameter(b.view(-1), requires_grad=True)
  82. def _print_biases(self):
  83. m = self.model[-1] # Detect() module
  84. for f in sorted([x % m.i for x in m.f]): #  from
  85. b = self.model[f].bias.detach().view(m.na, -1).T # conv.bias(255) to (3,85)
  86. print(('%g Conv2d.bias:' + '%10.3g' * 6) % (f, *b[:5].mean(1).tolist(), b[5:].mean()))
  87. def parse_model(md, ch): # model_dict, input_channels(3)
  88. print('\n%3s%15s%3s%10s %-40s%-30s' % ('', 'from', 'n', 'params', 'module', 'arguments'))
  89. anchors, nc, gd, gw = md['anchors'], md['nc'], md['depth_multiple'], md['width_multiple']
  90. na = (len(anchors[0]) // 2) # number of anchors
  91. no = na * (nc + 5) # number of outputs = anchors * (classes + 5)
  92. layers, save, c2 = [], [], ch[-1] # layers, savelist, ch out
  93. for i, (f, n, m, args) in enumerate(md['backbone'] + md['head']): # from, number, module, args
  94. m = eval(m) if isinstance(m, str) else m # eval strings
  95. for j, a in enumerate(args):
  96. try:
  97. args[j] = eval(a) if isinstance(a, str) else a # eval strings
  98. except:
  99. pass
  100. n = max(round(n * gd), 1) if n > 1 else n # depth gain
  101. if m in [nn.Conv2d, Conv, Bottleneck, SPP, DWConv, MixConv2d, Focus, ConvPlus, BottleneckCSP, BottleneckLight]:
  102. c1, c2 = ch[f], args[0]
  103. # Normal
  104. # if i > 0 and args[0] != no: # channel expansion factor
  105. # ex = 1.75 # exponential (default 2.0)
  106. # e = math.log(c2 / ch[1]) / math.log(2)
  107. # c2 = int(ch[1] * ex ** e)
  108. # if m != Focus:
  109. c2 = make_divisible(c2 * gw, 8) if c2 != no else c2
  110. # Experimental
  111. # if i > 0 and args[0] != no: # channel expansion factor
  112. # ex = 1 + gw # exponential (default 2.0)
  113. # ch1 = 32 # ch[1]
  114. # e = math.log(c2 / ch1) / math.log(2) # level 1-n
  115. # c2 = int(ch1 * ex ** e)
  116. # if m != Focus:
  117. # c2 = make_divisible(c2, 8) if c2 != no else c2
  118. args = [c1, c2, *args[1:]]
  119. if m is BottleneckCSP:
  120. args.insert(2, n)
  121. n = 1
  122. elif m is nn.BatchNorm2d:
  123. args = [ch[f]]
  124. elif m is Concat:
  125. c2 = sum([ch[x] for x in f])
  126. elif m is Origami:
  127. c2 = ch[f] * 5
  128. elif m is Detect:
  129. f = f or list(reversed([(-1 if j == i else j - 1) for j, x in enumerate(ch) if x == no]))
  130. else:
  131. c2 = ch[f]
  132. m_ = nn.Sequential(*[m(*args) for _ in range(n)]) if n > 1 else m(*args) # module
  133. t = str(m)[8:-2].replace('__main__.', '') # module type
  134. np = sum([x.numel() for x in m_.parameters()]) # number params
  135. m_.i, m_.f, m_.type, m_.np = i, f, t, np # attach index, 'from' index, type, number params
  136. print('%3s%15s%3s%10.0f %-40s%-30s' % (i, f, n, np, t, args)) # print
  137. save.extend(x % i for x in ([f] if isinstance(f, int) else f) if x != -1) # append to savelist
  138. layers.append(m_)
  139. ch.append(c2)
  140. return nn.Sequential(*layers), sorted(save), ch
  141. if __name__ == '__main__':
  142. parser = argparse.ArgumentParser()
  143. parser.add_argument('--cfg', type=str, default='yolov5s.yaml', help='model.yaml')
  144. parser.add_argument('--device', default='', help='cuda device, i.e. 0 or 0,1,2,3 or cpu')
  145. opt = parser.parse_args()
  146. opt.cfg = glob.glob('./**/' + opt.cfg, recursive=True)[0] # find file
  147. device = torch_utils.select_device(opt.device)
  148. # Create model
  149. model = Model(opt.cfg).to(device)
  150. model.train()
  151. # Profile
  152. # img = torch.rand(8 if torch.cuda.is_available() else 1, 3, 640, 640).to(device)
  153. # y = model(img, profile=True)
  154. # print([y[0].shape] + [x.shape for x in y[1]])
  155. # ONNX export
  156. # model.model[-1].export = True
  157. # torch.onnx.export(model, img, f.replace('.yaml', '.onnx'), verbose=True, opset_version=11)
  158. # Tensorboard
  159. # from torch.utils.tensorboard import SummaryWriter
  160. # tb_writer = SummaryWriter()
  161. # print("Run 'tensorboard --logdir=models/runs' to view tensorboard at http://localhost:6006/")
  162. # tb_writer.add_graph(model.model, img) # add model to tensorboard
  163. # tb_writer.add_image('test', img[0], dataformats='CWH') # add model to tensorboard