Using mobilenet_v3_edgetpu checkpoints

System information

• What is the top-level directory of the model you are using: research/slim/nets/mobilenet
• Have I written custom code (as opposed to using a stock example script provided in TensorFlow): yes
• OS Platform and Distribution (e.g., Linux Ubuntu 16.04): Linux Ubuntu 16.04
• TensorFlow installed from (source or binary): binary (pip)
• TensorFlow version (use command below): 1.14.0
• Bazel version (if compiling from source): -
• CUDA/cuDNN version: - (CPU)
• GPU model and memory: - (CPU)
• Exact command to reproduce:

Describe the problem

I am having trouble loading mobilenet_v3_edgetpu checkpoints and measuring top-1 accuracy.

1. for 1.0 version: I'm able to restore the checkpoint and run inference, but can't replicate the documented accuracy. I'm getting somewhere between 72%-73% vs. the expected over 75%.
   I thought the problem was not using the ExponentMovingAverage weights, but after fixing the code the accuracy further drops to about 64%.
2. for 0.75 version: I'm unable to load the checkpoint at all. I use the depth_multiplier arguments and set it to 0.75. When loading the checkpoint I get:

tensorflow.python.framework.errors_impl.InvalidArgumentError: Restoring from checkpoint failed. This is most likely due to a mismatch between the current graph and the graph from the checkpoint. Please ensure that you have not altered the graph expected based on the checkpoint. Original error:

Source code / logs

Code used to load checkpoints:
https://gist.github.com/shachargluska/76e187606a15d8310436525a791518da