[slim] Bug in replacing activation function in MobileNetV2

• What is the top-level directory of the model you are using: slim
• Have I written custom code: no
• OS Platform and Distribution (e.g., Linux Ubuntu 16.04): Ubuntu 16.04
• TensorFlow installed from (source or binary): binary
• TensorFlow version (use command below): 1.10.
• Bazel version (if compiling from source): -
• CUDA/cuDNN version: CUDA 9
• GPU model and memory: 1080Ti, 11GB
• Exact command to reproduce: Training MobileNetV2 with activation function other than ReLU6

It seems to me that there's a bug in here, where you edit the conv defaults of MNv2 to be with other activation function. While writing the conv defaults to conv_defaults, the variable used later is conv_defs. This results in no change of the activation function.