CRNN(
  11.77 M, 100.000% Params, 1.26 GMac, 100.000% MACs, 
  (layers): ModuleDict(
    11.77 M, 100.000% Params, 1.26 GMac, 100.000% MACs, 
    (conv1): Conv2d(640, 0.005% Params, 3.28 MMac, 0.260% MACs, 1, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    (relu1): ReLU(0, 0.000% Params, 327.68 KMac, 0.026% MACs, )
    (pool1): MaxPool2d(0, 0.000% Params, 327.68 KMac, 0.026% MACs, kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
    (conv2): Conv2d(73.86 k, 0.628% Params, 94.54 MMac, 7.494% MACs, 64, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    (relu2): ReLU(0, 0.000% Params, 163.84 KMac, 0.013% MACs, )
    (pool2): MaxPool2d(0, 0.000% Params, 163.84 KMac, 0.013% MACs, kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
    (conv3): Conv2d(295.17 k, 2.508% Params, 94.45 MMac, 7.487% MACs, 128, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    (bn3): BatchNorm2d(512, 0.004% Params, 163.84 KMac, 0.013% MACs, 256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
    (relu3): ReLU(0, 0.000% Params, 81.92 KMac, 0.006% MACs, )
    (conv4): Conv2d(590.08 k, 5.014% Params, 188.83 MMac, 14.968% MACs, 256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    (relu4): ReLU(0, 0.000% Params, 81.92 KMac, 0.006% MACs, )
    (pool4): MaxPool2d(0, 0.000% Params, 81.92 KMac, 0.006% MACs, kernel_size=(2, 2), stride=(2, 1), padding=(0, 1), dilation=1, ceil_mode=False)
    (conv5): Conv2d(1.18 M, 10.029% Params, 193.55 MMac, 15.342% MACs, 256, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    (bn5): BatchNorm2d(1.02 k, 0.009% Params, 167.94 KMac, 0.013% MACs, 512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
    (relu5): ReLU(0, 0.000% Params, 83.97 KMac, 0.007% MACs, )
    (conv6): Conv2d(2.36 M, 20.053% Params, 387.01 MMac, 30.677% MACs, 512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    (relu6): ReLU(0, 0.000% Params, 83.97 KMac, 0.007% MACs, )
    (pool6): MaxPool2d(0, 0.000% Params, 83.97 KMac, 0.007% MACs, kernel_size=(2, 2), stride=(2, 1), padding=(0, 1), dilation=1, ceil_mode=False)
    (conv7): Conv2d(1.05 M, 8.915% Params, 43.01 MMac, 3.409% MACs, 512, 512, kernel_size=(2, 2), stride=(1, 1))
    (bn7): BatchNorm2d(1.02 k, 0.009% Params, 41.98 KMac, 0.003% MACs, 512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
    (relu7): ReLU(0, 0.000% Params, 20.99 KMac, 0.002% MACs, )
    (lstm1): LSTM(1.58 M, 13.401% Params, 64.87 MMac, 5.142% MACs, 512, 256, bidirectional=True)
    (fc1): Linear(131.33 k, 1.116% Params, 5.37 MMac, 0.426% MACs, in_features=512, out_features=256, bias=True)
    (lstm2): LSTM(1.05 M, 8.945% Params, 43.37 MMac, 3.438% MACs, 256, 256, bidirectional=True)
    (fc2): Linear(3.46 M, 29.364% Params, 141.41 MMac, 11.209% MACs, in_features=512, out_features=6736, bias=True)
  )
)