How could i see the forward process of certain model?

[xanhug]

Now i'm learning from a paper,but it's code is encapsulated in timm,named interfuser.
when i try to definite this model and print it,i can only get information about this model's structure like this

Interfuser(
  (rgb_backbone): FeatureListNet(
    (conv1): Sequential(
      (0): Conv2d(3, 32, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)
      (1): BatchNorm2d(32, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (2): ReLU(inplace=True)
      (3): Conv2d(32, 32, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
      (4): BatchNorm2d(32, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (5): ReLU(inplace=True)
      (6): Conv2d(32, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
    )
    (bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
    (act1): ReLU(inplace=True)
    (maxpool): MaxPool2d(kernel_size=3, stride=2, padding=1, dilation=1, ceil_mode=False)
    (layer1): Sequential(

but i want to figure out the forward procession of this model,is there exist any command or something help to print the forward procession of a model like

    def forward(self, image, lidar, velocity):
        '''
        Image + LiDAR feature fusion using transformers
        Args:
            image_list (list): list of input images
            lidar_list (list): list of input LiDAR BEV
            velocity (tensor): input velocity from speedometer
        '''

        if self.image_encoder.normalize:
            image_tensor = normalize_imagenet(image)
        else:
            image_tensor = image

        lidar_tensor = lidar

        image_features = self.image_encoder.features.conv1(image_tensor)
        image_features = self.image_encoder.features.bn1(image_features)
        image_features = self.image_encoder.features.act1(image_features)
        image_features = self.image_encoder.features.maxpool(image_features)
        lidar_features = self.lidar_encoder._model.conv1(lidar_tensor)
        lidar_features = self.lidar_encoder._model.bn1(lidar_features)
        lidar_features = self.lidar_encoder._model.act1(lidar_features)
        lidar_features = self.lidar_encoder._model.maxpool(lidar_features)

        image_features = self.image_encoder.features.layer1(image_features)
        lidar_features = self.lidar_encoder._model.layer1(lidar_features)

This way I can figure out which features are handed to which module,and if there are any additional operation like concatenate along the way.
thanks for your kind help.

[hankyul2]

Hi @xanhug

How about using torchsummary? This package is able to sequentially lists the hidden layer name with the output shape along with the forward process.

----------------------------------------------------------------
        Layer (type)               Output Shape         Param #
================================================================
            Conv2d-1           [-1, 10, 24, 24]             260
            Conv2d-2             [-1, 20, 8, 8]           5,020
         Dropout2d-3             [-1, 20, 8, 8]               0
            Linear-4                   [-1, 50]          16,050
            Linear-5                   [-1, 10]             510
================================================================
Total params: 21,840
Trainable params: 21,840
Non-trainable params: 0
----------------------------------------------------------------
Input size (MB): 0.00
Forward/backward pass size (MB): 0.06
Params size (MB): 0.08
Estimated Total Size (MB): 0.15
----------------------------------------------------------------

Thank you.

hankyul

[xanhug]

thanks for your help,I have found the forward function in a .py file under timm/models