kingxueyuf / predict-car-speed Goto Github PK

[!] Model Summary:
AlexLSTM (
  (conv): Sequential (
    (0): Sequential (
      (0): Conv2d(3, 64, kernel_size=(11, 11), stride=(4, 4), padding=(2, 2))
      (1): ReLU (inplace)
      (2): MaxPool2d (size=(3, 3), stride=(2, 2), dilation=(1, 1))
      (3): Conv2d(64, 192, kernel_size=(5, 5), stride=(1, 1), padding=(2, 2))
      (4): ReLU (inplace)
      (5): MaxPool2d (size=(3, 3), stride=(2, 2), dilation=(1, 1))
      (6): Conv2d(192, 384, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
      (7): ReLU (inplace)
      (8): Conv2d(384, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
      (9): ReLU (inplace)
      (10): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
      (11): ReLU (inplace)
      (12): MaxPool2d (size=(3, 3), stride=(2, 2), dilation=(1, 1))
    )
  )
  (lstm): LSTM(12288, 1600, num_layers=3, dropout=0.3)
  (fc): Sequential (
    (0): Linear (1600 -> 512)
    (1): ReLU ()
    (2): Dropout (p = 0.2)
    (3): Linear (512 -> 64)
    (4): ReLU ()
    (5): Dropout (p = 0.2)
    (6): Linear (64 -> 1)
  )
)

CNN Layer:
-> (batch_size, 20, hh, ww, 3) 
-> AlexNet -> (batch_size, 20, hh', ww', filter_size) 
-> flatten -> (batch_size, 20, 12288)

LSTM Layer: 
# 20 images per batch
# get the last 19 cells output (LSTM[1:]) from the last layer of LSTM
# as we can't predict speed for the 1st image because it is lack of context (no image before the 1st one)
-> (batch_size, 20, 12288) 
-> LSTM -> (batch_size, 19, 1600)  

Fully Connected Layer:
-> (batch_size, 19, 1600) 
-> fc_layer_1 -> (batch_size, 19, 512) 
-> fc_layer_2 -> (batch_size, 19, 64) 
-> fc_layer_3 -> (batch_size, 19, 1)