MobilenetV1 difference from original paper

[gussmith23]

Others (#35 #53) have noticed discrepancies in model architectures as well. Just noting a discrepancy in MobilenetV1: https://github.com/kuangliu/pytorch-cifar/blob/ab908327d44bf9b1d22cd333a4466e85083d3f21/models/mobilenet.py#L32 I'm fairly certain that the first convolution should have stride of 2! Though when I make this change, it results in lowered accuracy overall.

[renebidart]

CIFAR images are only 32x32 (compared to imagenet 224x224), so you need to reduce the stride of the first few layers, or else they will perform poorly.