I’m passing in an alexnet model from the pretrainedmodels repository to create_cnn, using the default alexnet head rather than adaptive pooling.
Does anyone know why the summary information is incorrect although the model information is correct and the model seems to train properly? I’m having a bit of trouble following the how the summary code works here. (using the dogs/cats breed dataset that has 37 classes). For example:
import pretrainedmodels
data = ImageDataBunch.from_name_re(path_img, fnames, pat, ds_tfms=get_transforms(), size=224, bs=64
).normalize(imagenet_stats)
arch = pretrainedmodels.__dict__['alexnet'](num_classes=1000,pretrained='imagenet')
arch.last_linear.out_features = data.c
model = nn.Sequential(OrderedDict([
('features', arch._features),
('classifier', nn.Sequential(Flatten(), *children(arch)[1:]))
]))
learn = create_cnn(data, lambda *args : model, metrics=error_rate, custom_head=net.classifier)
# model says 37 classes in last linear
print(learn.model)
# summary incorrectly says 1000 classes in last linear
print(learn.summary())
Output:
Sequential(
(0): Sequential(
(0): Sequential(
(0): Conv2d(3, 64, kernel_size=(11, 11), stride=(4, 4), padding=(2, 2))
(1): ReLU(inplace)
(2): MaxPool2d(kernel_size=3, stride=2, padding=0, dilation=1, ceil_mode=False)
(3): Conv2d(64, 192, kernel_size=(5, 5), stride=(1, 1), padding=(2, 2))
(4): ReLU(inplace)
(5): MaxPool2d(kernel_size=3, stride=2, padding=0, dilation=1, ceil_mode=False)
(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(kernel_size=3, stride=2, padding=0, dilation=1, ceil_mode=False)
)
)
(1): Sequential(
(0): Flatten()
(1): Dropout(p=0.5)
(2): Linear(in_features=9216, out_features=4096, bias=True)
(3): ReLU(inplace)
(4): Dropout(p=0.5)
(5): Linear(in_features=4096, out_features=4096, bias=True)
(6): ReLU(inplace)
(7): Linear(in_features=4096, out_features=37, bias=True)
)
)
======================================================================
Layer (type) Output Shape Param # Trainable
======================================================================
Conv2d [1, 64, 55, 55] 23,296 False
______________________________________________________________________
ReLU [1, 64, 55, 55] 0 False
______________________________________________________________________
MaxPool2d [1, 64, 27, 27] 0 False
______________________________________________________________________
Conv2d [1, 192, 27, 27] 307,392 False
______________________________________________________________________
ReLU [1, 192, 27, 27] 0 False
______________________________________________________________________
MaxPool2d [1, 192, 13, 13] 0 False
______________________________________________________________________
Conv2d [1, 384, 13, 13] 663,936 False
______________________________________________________________________
ReLU [1, 384, 13, 13] 0 False
______________________________________________________________________
Conv2d [1, 256, 13, 13] 884,992 False
______________________________________________________________________
ReLU [1, 256, 13, 13] 0 False
______________________________________________________________________
Conv2d [1, 256, 13, 13] 590,080 False
______________________________________________________________________
ReLU [1, 256, 13, 13] 0 False
______________________________________________________________________
MaxPool2d [1, 256, 6, 6] 0 False
______________________________________________________________________
Flatten [1, 9216] 0 False
______________________________________________________________________
Dropout [1, 9216] 0 False
______________________________________________________________________
Linear [1, 4096] 37,752,832 True
______________________________________________________________________
ReLU [1, 4096] 0 False
______________________________________________________________________
Dropout [1, 4096] 0 False
______________________________________________________________________
Linear [1, 4096] 16,781,312 True
______________________________________________________________________
ReLU [1, 4096] 0 False
______________________________________________________________________
Linear [1, 1000] 4,097,000 True
______________________________________________________________________
Total params: 61,100,840
Total trainable params: 58,631,144
Total non-trainable params: 2,469,696