Inceptionv3 layers

[firasl]

Hi,
which pre-trained conv2d layers from inceptionv3 did you use?
When I compile your model, I find that the first three convolutional layers have weights sizes of (3,3,32,32), (3,3,32,64), (3,3,3,16). While the first Conv weights of inceptionV3 have sizes (3,3,3,32),(3,3,32,32),(3,3,32,64). So the weights do not match.